Written by Zhang — Paddy & Water Inc. / Data & Analytics

Design, Quality, Maintenance, Project Management, Core-System Integration, and the DX Roadmap

This report systematically explains the use of AI in engineering work (product design, quality assurance, equipment maintenance, software development, and project management), covering concrete technologies, implementation patterns, integration cases with core enterprise systems (ERP/PLM/MES), and measured results. Based on the latest 2025-2026 case studies and survey data, it is intended for use as a reference document for DX promotion in manufacturing engineering departments.

Chapter 1: The Overall Framework of Engineering DX

1-1 The Digital Thread: The Core Concept of Engineering DX

The Digital Thread is the core concept of engineering DX, referring to “a continuous flow of digital data spanning the entire product lifecycle, from design through manufacturing, use, and disposal.” The ideal state is one in which PLM (Product Lifecycle Management), CAD, ERP, MES, and IoT data are seamlessly linked, so that design changes are reflected in real time in production plans, quality records, and parts inventory.

Layer	System	Managed Data	Role of AI
Design & Development	CAD/CAE, PLM (Windchill/Teamcenter)	Drawings, BOM, specifications, design change history, CAE simulation results	Generative Design, AI CAE surrogate models, design optimization proposals
Manufacturing	ERP (SAP S/4HANA), MES, schedulers	Production instructions, bill of materials (BOM), process routing, cost	AI production scheduling, bottleneck prediction, yield prediction
Quality	QMS, MES quality modules, visual inspection AI	Defect records, inspection results, process capability (Cpk), FMEA data	AI image inspection, SPC anomaly detection, quality risk prediction, defect root-cause analysis
Maintenance	EAM (SAP PM), CMMS, IoT platforms	Equipment master data, maintenance history, sensor data (vibration/temperature/current)	AI predictive maintenance, failure probability prediction, optimal maintenance scheduling
Project Management	ERP PS (SAP PS/P&RM), PPM, MS Project	WBS, actual man-hours, budget, resource allocation, milestones	AI delay prediction, risk scoring, optimal resource allocation, EVM automation

※ Implementing a digital thread takes a lead time of 18 to 36 months (the median across multiple implementation cases). Data standardization, building an API ecosystem across systems, and change management are the largest investment items. A small start—beginning with a single data flow pair such as “PLM ↔ ERP BOM integration”—is recommended.

1-2 The AI Maturity Model: Where Engineering Departments Stand Today

The AI maturity of the PLM/engineering field can be organized into a four-level model (Engineering.com’s 2025 framework).

• Level 1 (Automation): Automated design checks, automatic part-number assignment, and standard-parts recommendations. As of the end of 2025, almost all major manufacturers have implemented or are implementing this level

• Level 2 (Prediction and Optimization): AI-driven function prediction, design optimization, reliability prediction, and defect risk prediction. Full-scale rollout is underway from 2025 onward alongside digital thread development

• Level 3 (Autonomous Design Support): AI-led design proposal generation through Generative Design, AI acceleration of CAE simulation, and material-selection AI. Leading companies are implementing this level

• Level 4 (Fully AI-Driven Development): AI interprets design requirements and autonomously generates and verifies everything from the initial concept through design for manufacturability (DFM). Widespread practical adoption is predicted from 2030 onward

Chapter 2: The Use of AI in Design and Development

2-1 Generative Design

Generative Design is a CAD function in which a designer inputs “constraints” (materials, loads, cost, manufacturing method) and “optimization goals” (weight reduction, maximum stiffness, minimum cost), and the AI automatically generates thousands to tens of thousands of design candidates and presents the optimal solution. Autodesk Fusion 360, PTC Creo, and Siemens NX are the leading products.

• Application example: lightweight design of an aircraft bracket part. Manual design (repeated trial and error taking 4-8 weeks) was shortened to 2-3 days using Generative Design. The AI-generated shape took on an organic structure resembling a biological skeleton, cutting weight by 40-60% compared with the conventional design while maintaining equivalent stiffness

• Integration with manufacturing constraints: the AI incorporates manufacturing methods such as 3D printing, machining, and casting as constraints, so that it is designed to “never propose a shape that cannot be manufactured.” DFM (Design for Manufacturing) is thereby integrated with AI

[Case study] Automotive parts manufacturer (Autodesk Generative Design): lightweight design of an EV suspension bracket. Achieved a 32% weight reduction versus the conventional design. Shortened design study man-hours from three weeks to four days. At the same time, the AI automatically generated shape variations that satisfied seven manufacturing constraints (casting, machining, welding)

2-2 AI-CAE Surrogate Models (Accelerating Simulation)

CAE (Computer Aided Engineering) simulation (FEM analysis, CFD analysis) is essential for verifying design quality, but high-precision analysis takes hours to days of computation time, which has become a bottleneck for design iteration. AI surrogate models solve this problem.

• Mechanism: a deep learning model is trained on a large volume of past CAE analysis results (pairs of design parameters and analysis results). Once trained, the model outputs a “prediction equivalent to the analysis result” in milliseconds when new design parameters are input

• Accuracy and use cases: 95-99% accuracy (versus full FEM analysis). Used in the design-exploration phase (rapidly screening a large number of design candidates), so that only promising design candidates are subjected to full CAE analysis, greatly reducing overall analysis man-hours

[Case study] Electronics manufacturer (thermal design): introduced an AI surrogate model for CFD analysis of board heat-dissipation design. Shortened the prediction time for design parameters (fin size, placement) → heat-dissipation performance from four hours to 0.3 seconds. Expanded the number of design-exploration candidates 100-fold and improved the heat-dissipation performance of the final adopted design by 15%

2-3 AI Use Through PLM and ERP BOM Integration

Mismatches and synchronization delays between the design BOM (Engineering BOM: E-BOM) on PLM and the manufacturing BOM (Manufacturing BOM: M-BOM) on ERP are a chronic problem at many manufacturers. AI strengthens this integration.

Core-System Data AI Use Case: Automatic E-BOM → M-BOM Conversion

• Challenge: the E-BOM is managed in PLM (the design definition of a part), while the M-BOM is managed in ERP (manufacturing specifications including production routing and process breakdown). Manual E-BOM → M-BOM conversion work occurs with every design change. Conversion errors and synchronization delays are a cause of parts mix-ups on the shop floor

• The AI solution: the AI learns past E-BOM → M-BOM conversion patterns and automatically generates “proposed changes to the M-BOM” whenever a design change occurs. The AI also judges exception patterns in the conversion rules (which process BOM is affected by a given design change)

• Effect: at a major industrial equipment manufacturer (S/4HANA plus Teamcenter PLM integration), the time to reflect design changes was shortened from an average of five days to within one day. Manufacturing defects caused by conversion errors were reduced by 80%

Engineering Change (ECO/ECN) Management and the Use of ERP Data

• ECO (Engineering Change Order) impact analysis: the AI automatically identifies, across PLM and ERP inventory data, every product, process, inventory item, and already-ordered part affected by a design change, presenting information such as “245 SKUs are affected by this change, with an inventory-obsolescence risk of 28 million yen” before the design change is approved

• Prioritizing changes: the AI cross-references ERP order and production-plan data with PLM change requests to help judge “does this change need to be made right now, or can it wait until the next model change,” including the economic impact

Chapter 3: Quality Assurance AI

3-1 AI Image Inspection (Detailed Implementation)

Inspection System Configuration

• Hardware: line cameras (area cameras, line-scan cameras) plus lighting (dome, backlight, laser) plus edge computing devices (NVIDIA Jetson, Intel NUC, etc.)

• Software: deep learning (CNNs such as ResNet and EfficientDet) or anomaly detection (an autoencoder approach that learns only from normal products to detect anomalies). Transfer learning enables high accuracy even with a small number of defect samples

• Integration with MES: when the AI makes an “NG” determination, defect data (defect type, location, destination) is sent to MES in real time. Automatic sorting of defective products, a stop signal to the processing line, and automatic generation of quality records are executed in an integrated fashion

Inspection Target	Typical AI Detection Accuracy	Comparison with Manual Inspection	Approximate Implementation Cost
Surface scratches/flaws (metal, resin)	99.3-99.8%	Manual: 92-95% (declines with fatigue)	Full edge-AI setup: 5-20 million yen
Dimension/shape measurement	±0.01-0.1mm (depends on camera resolution)	Equal to or better than a projection inspection machine; more than 10x the throughput	Line camera + measurement AI: 3-15 million yen
Label/print inspection	99.9% or higher	Manual inspection has many oversights (checks the full run in a fraction of a second per unit)	Relatively low cost (1-5 million yen)
Electronic boards (SMT mounting)	98.5-99.5% (AI-enabled AOI)	50-70% reduction in false-call rate versus conventional AOI	AI upgrade of existing AOI: 2-8 million yen

3-2 Advancing SPC (Statistical Process Control) with AI

SPC (Statistical Process Control) is a method for statistically managing the quality of a manufacturing process, but conventional control charts (X-bar/R charts, etc.) have remained at reactive detection—”detecting an anomaly that has already occurred.” Integrating AI with SPC realizes predictive quality control.

• Multivariate SPC: whereas conventional SPC managed one parameter at a time, AI simultaneously monitors the correlations among multiple process parameters. The AI detects subtle process drift where “each parameter individually is within the normal range, but the combined pattern across multiple parameters is abnormal”

• Root-cause analysis of anomalies: when a control deviation occurs, the AI cross-references past similar patterns and their root causes (tool wear, material lot change, ambient temperature change, operator change) to present candidate causes—automating 4M analysis

[Case study] Semiconductor manufacturing equipment maker (ERP data x SPC AI): monitors 50 parameters of a mass-production process with multivariate AI-SPC. A process that previously “investigated the cause after a defective product was produced” was transformed into one that “detects process drift before a defect occurs and issues automatic adjustment instructions.” Process defect rate was reduced from 0.42% to 0.18% within six months

Chapter 4: Equipment Maintenance AI (In Detail)

4-1 Predictive Maintenance Roadmap by Implementation Level

Maturity Level	Name	Description	Required Systems
Level 1	Reactive Maintenance (Break-Fix)	Fix it when it breaks. Highest cost (emergency response, production stoppage)	None (luck)
Level 2	Time-Based Preventive Maintenance (Time-based PM)	Calendar-based periodic inspection and parts replacement. Low cost-effectiveness (replaces parts that are still usable)	CMMS (maintenance management system) only
Level 3	Condition-Based Maintenance	Alerts when sensor monitoring (vibration, temperature) exceeds a threshold. No predictive AI	IoT sensors + alert system
Level 4	AI Predictive Maintenance	AI learns degradation patterns and forecasts, as a probability, “how many hours until failure”	IoT + machine learning model + EAM integration
Level 5	AI Autonomous Maintenance (Prescriptive)	AI autonomously judges and issues instructions for the optimal maintenance timing, maintenance work content, and even spare-parts procurement	Level 4 + ERP MM/PM integration + maintenance planning AI

4-2 Integrating Core Systems (SAP PM/EAM) with AI Predictive Maintenance

To maximize the effectiveness of AI predictive maintenance, it is important to integrate IoT sensor data with EAM (Enterprise Asset Management) / SAP PM (Plant Maintenance) data. Sensor data alone does not reveal an asset’s usage history, parts-replacement records, or past failure causes, which limits prediction accuracy.

Specific Patterns for Integrating SAP PM Data with AI

• Integrating equipment master data with sensor data: the AI integrates SAP PM equipment master data (equipment number, installation date, specifications, date of last overhaul) with real-time IoT sensor data, building a multivariate failure-prediction model that combines “X years since installation, Y hours of operation since the last replacement, and the current vibration value Z”

• Learning from maintenance history: the AI learns from past SAP PM maintenance orders (failure content, replaced parts, repair time, cost), learning patterns such as “this equipment’s bearings show signs of degradation after an average of X hours,” and generating a prediction model optimized for each individual asset

• Integration with spare-parts inventory: the AI predictive-maintenance forecast “a part will need replacement in X days” is automatically linked to SAP MM (inventory management). When spare-parts inventory is insufficient, an automatic purchase-request is generated, eliminating the situation of “no parts on hand after the breakdown”

• Automatic optimization of maintenance planning: predictive-maintenance forecasts for multiple assets are integrated into the SAP PS maintenance plan. The AI proposes an optimal schedule for “which equipment to maintain when, by whom, and for how many hours” by cross-referencing it against plant-wide resources (technician availability, equipment downtime schedule)

[Case study] Steel plant (SAP PM + IoT + AI integration): built an AI model integrating IoT sensor data from 200 blast-furnace and rolling-mill assets with SAP PM maintenance history. Unplanned stoppages were reduced by 42% year over year. Annual emergency maintenance costs were reduced by 320 million yen. By detecting the optimal timing for a single planned maintenance activity an average of 2.3 days earlier, the AI converted emergency stoppages into planned stoppages

[Case study] Chemical plant (SAP PM + predictive maintenance AI): introduced AI predictive maintenance for 120 pumps and compressors. Maintenance costs were reduced by 25% (eliminating unnecessary periodic replacements and reducing emergency-response costs). Spare-parts inventory was simultaneously reduced by 15% (just-in-time procurement based on AI predictions)

Chapter 5: Advancing Engineering Project Management with AI

5-1 The Distinct Nature of Project Management in Manufacturing

Engineering projects in manufacturing (new product development, new equipment installation, plant relocation, large-scale renovation) carry a complexity different from consulting or IT projects. Dependencies spanning multiple departments (design, procurement, manufacturing preparation, quality), physical constraints (prototypes, equipment lead times, construction schedules), and linkage with core ERP systems (materials ordering, cost management) are all intertwined at once.

5-2 AI Project Management Using ERP (SAP PS) Data

Main Patterns for Using SAP PS Data x AI

• Automatic analysis of the gap between actual man-hours and the plan: automatically compares actual man-hours by WBS element (ATH: Actual Time Recorded) in SAP PS against planned man-hours (PTD: Planned Time to Date). AI deviation analysis calculates “given the current burn rate, what is the probability the project will finish on time”

• AI delay prediction (schedule risk): the AI performs an integrated analysis of dependencies among WBS elements, remaining work for each task, resource utilization rates, and patterns of similar delays from past projects, scoring tasks on the critical path with high delay risk to provide early warning

• Automating EVM (Earned Value Management): automatically calculates CPI (Cost Performance Index), SPI (Schedule Performance Index), and EAC (Estimate at Completion) from SAP PS cost data. The AI presents “the completion date and completion cost if this trend continues” as a probability distribution

• Optimal resource allocation: when multiple projects run in parallel, the AI optimizes the allocation of engineers, equipment, and outsourced resources. The AI automatically executes SAP PS resource planning and capacity leveling

AI Application Area	SAP Data Used	AI Function	Expected Effect
Progress Management	WBS actual man-hours (PS), confirmations (CNF), milestones	Completion-probability prediction, delay alerts, automatic critical-path updates	Improved early detection rate of project delays; reduced PM reporting workload
Cost Management	WBS cost plan/actuals (PS/CO), commitments	Automatic EAC calculation, cost-overrun risk prediction, additional-order prediction	Improved final-cost prediction accuracy; proactive suppression of cost overruns
Resource Management	Capacity (PS/PP), utilization actuals (HR or PS)	Bottleneck resource detection, optimal assignment proposals	10-15% improvement in engineer utilization; reduced idle time
Procurement/Materials Management	Materials orders (MM), procurement lead time, project BOM	Prediction of procurement delay risk for critical materials, alternative-part proposals	Reduced project wait time due to material shortages

Integration Patterns Between AI Project Management Tools and SAP PS

• SAP Portfolio and Project Management (formerly PPM): SAP’s standard portfolio-management tool. AI functions (delay prediction, resource optimization) were added in the 2025 update. It natively references S/4HANA PS data

• Microsoft Project + Copilot: Microsoft Project’s AI functions (Copilot in Project) provide schedule analysis, risk identification, and automatic progress-report generation. Integration with SAP PS is possible via the Microsoft Power Platform

• Primavera P6 (for construction/plant): geared toward large-scale infrastructure and construction projects. Standard equipped with AI-driven Monte Carlo simulation (completion-probability analysis). Has a track record of integration with SAP PS

[Case study] A large plant EPC (engineering, procurement, construction) firm: introduced SAP PS x AI delay prediction. An AI monitoring a WBS of 1,200 tasks automatically identifies the “top 20 delay-risk tasks” every week and provides the PM with a prioritized action list. The project completion rate (on time and on budget) improved from 54% before introduction to 73%

[Case study] A major manufacturer’s new plant construction project (SAP PS + AI): introduced AI-EVM for a new plant construction project (WBS with 680 elements, a budget of 30 billion yen). Monthly EAC predictions reached an accuracy within ±3% (versus ±15% previously). Six months before project completion, the AI detected “signs of a budget overrun,” and the overrun was avoided through scope reduction and design optimization

5-3 Digital Twins and Project Simulation

Digital twins are also increasingly applied to project management. By simulating layout changes, equipment additions, and logistics flows on a 3D digital twin model of a plant, it becomes possible to discover and resolve in advance “problems that would otherwise only be noticed after actual construction and installation.”

• Plant layout simulation: using simulators such as Emulate3D and Tecnomatix Plant Simulation, the logistics flow, utilization rate, and throughput of new line additions and equipment layout changes are simulated. ERP production-plan and logistics data are loaded into the simulator to conduct verification under realistic conditions

• Construction BIM (Building Information Modeling) x AI scheduling: in construction and plant work, process simulation is performed using 4D-BIM (a 3D construction model plus a time axis). The AI automatically optimizes crane movement paths, the sequence of material delivery, and interference among work areas, reducing rework on actual construction sites

※ The essential value of digital twin x AI project simulation lies in “digitally testing, in advance, risks that would otherwise only be known by trying.” It is a high-ROI area that can greatly reduce the “cost of failure” in design, construction, and project planning.

5-4 Automating PMO Work with Generative AI

Among the applications of AI project management, the one with the lowest barrier to adoption and the fastest payoff is “automating PMO administrative work with generative AI.” Project managers and PMOs are said to spend 30-40% of their working hours on “coordination and administration”—progress aggregation, report writing, meeting minutes, and issue management. Generative AI substantially compresses this workload, shifting the PM’s role from “report writer” to “decision-maker.”

• Automatic generation of progress reports: generative AI reads the actuals from SAP PS, schedule-management tools, and timesheets, and automatically drafts a weekly or monthly progress summary (completion rate, causes of delay, next week’s actions). The PM only needs to review and correct it.

• Meeting minutes and action-item extraction: generative AI creates minutes from the recording/transcript of a regular meeting, structures the decisions and action items on “who will do what, by when,” and automatically registers them in the issue-management log.

• Automatic updates to the risk register: the AI monitors entries in the issue-management log, chat, and email, detects signs of a new risk (statements such as “the materials won’t arrive in time”), and automatically logs it in the risk list with a severity score.

• Stakeholder-specific summaries: generative AI automatically rewrites the same progress information differently depending on the audience—a one-page summary for executives versus detailed tasks for the field. In multilingual projects, translation is performed at the same time.

PMO Task	Conventional Time Required	After Applying Generative AI	Main Effect
Weekly progress report creation	Half a day per week	30 minutes (review only)	80% reduction in creation effort
Regular meeting minutes/issue logging	1-2 hours per meeting	10 minutes (review)	Prevents omissions; immediate sharing
Risk inventory review	Several hours monthly	Continuous automatic monitoring	Earlier detection by 2-4 weeks
Management reporting materials	1 day per month	2 hours	More focus on decision-making

※ Automating PMO work with generative AI is not a measure in which “AI replaces human judgment,” but rather one in which “AI increases the time people have available for judgment.” Starting here, acclimating the organization to working with AI, and then moving on to the predictive AI described in the next section is the realistic order of progression.

5-5 Specialized Tools and Implementation of Schedule-Risk Prediction AI

In addition to general-purpose project management tools, in construction, plant, and large-scale equipment projects, the implementation of specialized tools that “generate, optimize, and assess the risk of the schedule itself using AI” is advancing. These tools are trained on hundreds of thousands of past project records and are distinguished by their ability to quantify process planning that had previously relied on human experience.

Major Schedule-Risk AI Tools

• Generative scheduling type (represented by ALICE): automatically generates a 4D schedule that accounts for construction sequence and resource constraints from a 3D model (BIM) or block diagram. It explores millions of process scenarios and presents the plan with the minimum duration and cost. It dynamically re-optimizes in response to changing site conditions.

• Schedule-risk prediction type (trained on past actuals): an AI trained on hundreds of thousands of real project schedules presents, as a probability, “this task has a high likelihood of running later than planned” for a schedule being created. It surfaces “hidden delay risk” that does not appear on the critical path.

• Monte Carlo simulation type (Primavera P6, etc.): the duration of each task is entered as a probability distribution, and thousands of trials calculate the “probability distribution of the completion date (P50/P80).” Used as a standard in large infrastructure and EPC projects.

• AI-assistant type for general-purpose PM tools (Microsoft Project Copilot, etc.): supports schedule analysis, risk identification, and progress-report generation through natural language. Highly compatible with small and mid-sized projects and in-house standard tools.

Tool Type	Input Data	Core AI Function	Representative Application Area
Generative Scheduling	3D/BIM, block diagrams	Automatic schedule generation, dynamic optimization	Construction, plants, data centers
Trained on Past Actuals, Delay Prediction	Hundreds of thousands of past processes	Delay-probability scoring	Infrastructure, large-scale equipment
Monte Carlo	Probability distribution of task duration	Completion probability distribution (P50/P80)	EPC, public works
AI Assistant	Existing schedule	Analysis, reporting, recommendations	General manufacturing, in-house projects

[Case study] Generative scheduling on a large construction project: the AI automatically generated the construction schedule from a 3D-BIM model and compared multiple construction scenarios. It shortened the construction period by about 17%, and reduced labor costs by about 14% and equipment costs by about 12%. In some data center construction programs, schedules were compressed by up to 40%.

[Case study] Delay prediction by an AI trained on past actuals: an AI trained on more than 750,000 real project schedules was applied to a large civil-engineering project. Before construction began, the AI identified the “work sections with a high probability of completion delay,” and by building countermeasures into the plan at the planning stage, schedule overruns were reduced by 17-30% compared with the conventional method.

※ The essence of schedule-risk AI lies in “giving a probabilistic yardstick to the schedule lines that people previously drew from experience and intuition.” Its greatest value is being able to detect overly optimistic plans before construction begins and to take action before a delay materializes.

5-6 AI-Driven Project Portfolio Management

Beyond individual project management, AI adoption is also advancing in “portfolio management,” which optimizes across multiple projects. In manufacturing, dozens of new product development, capital investment, and improvement projects run simultaneously, and how to allocate limited engineers, prototyping equipment, and budget among them is a management-level challenge. AI supports this overall optimization.

• Automatic detection of resource conflicts: the AI cross-references the staffing plans of multiple projects and automatically flags overloads and conflicts, such as “the same engineer is assigned to three projects simultaneously next month.”

• Portfolio prioritization: the AI scores each project’s strategic fit, ROI, risk, and progress, visualizing which projects “should continue to receive investment” versus “should be scaled back or discontinued,” and providing input for gate reviews.

• What-if scenario analysis: the AI simulates questions such as “if the flagship project is moved up by two months, what is the impact on other projects’ resources,” allowing the impact to be quantitatively evaluated before a management decision is made.

• Automatic extraction of cross-project knowledge: the AI analyzes the causes of delays and cost overruns in past similar projects and presents “failure patterns that occurred in similar past projects” as a warning when planning a new project.

[Case study] Resource optimization across multiple development projects: a manufacturer that runs around 30 new-product-development projects in parallel at any given time introduced AI portfolio management. The AI monitors engineer utilization across projects, and by leveling overloads, the on-time delivery rate for key development projects improved and the chronic concentration of tasks on certain engineers was resolved.

5-7 AI Project Management: An Implementation Roadmap and Points to Consider

Success with AI project management does not come from “suddenly introducing delay-prediction AI,” but requires a staged rollout accompanied by data preparation and organizational familiarization. Below is a realistic approach assuming a manufacturing engineering department.

Stage	Main Initiatives	Prerequisite Data	Value Obtained
Stage 1: Visualization	Consolidating progress, man-hour, and cost data	SAP PS and timesheet setup	Quantitative grasp of the current state
Stage 2: Automation	Automating reports and minutes with generative AI	Digitization of routine data	Reduced administrative workload
Stage 3: Prediction	Predicting delays and cost overruns with AI	Accumulation of past project actuals	Early warning; proactive countermeasures
Stage 4: Optimization	Optimal allocation across portfolio and resources	Company-wide project data integration	Management-level overall optimization

• The data-quality wall: the AI’s prediction accuracy depends entirely on the quality of the input data. If the granularity of the WBS, the accuracy of man-hour entries, and the records of past projects are inconsistent, the AI cannot produce reliable predictions. “Prepare the data before introducing AI” is the golden rule.

• Buy-in from the field (change management): even if the AI warns “this task will be late,” it is meaningless if the field does not accept it, thinking “what does AI know.” It is important to present the basis for the prediction, initially position it as supporting human judgment, and build trust by accumulating a track record of accuracy.

• Avoiding overreliance: AI predictions are ultimately probabilistic and cannot fully account for physical constraints or sudden events (design changes, disasters, specification changes). The final decision must always be made by a human, with AI positioned as “a tool that reduces oversights.”

※ The key to maximizing the ROI of AI project management lies not in “the accuracy of the predictive AI” but in “data preparation” and “organizational mastery of the tools.” Building the mechanisms and culture to correctly accumulate project data before the technology is what appears to be a detour but is actually the shortest path.

Chapter 6: Software Development AI and the Engineering Organization

6-1 AI-Assisted Software Development (Embedded and Industrial Software)

AI is also transforming the software developed and maintained by engineering departments, including product-embedded software, plant control software, MES, and quality systems.

• AI coding assistants: AI coding assistants such as GitHub Copilot, JetBrains AI, and Tabnine are now widespread in industrial software development as well, providing code completion, test generation, automatic documentation generation, and code-review support. Cases report a 30-55% increase in development speed

• Legacy code analysis: AI automatically analyzes decades-old Fortran/COBOL/C manufacturing control programs and reverse-engineers specification documents from them, supporting the visualization and phased modernization of legacy systems that had been “black boxes”

[Case study] The engineering division of an automotive manufacturer: company-wide rollout of GitHub Copilot Enterprise (2025). Deployed to 800 embedded software developers. Reduced the effort required for unit-test generation and code review. In a developer survey, respondents reported that “coding speed increased by an average of 35%”

6-2 The DX Roadmap for an AI-Driven Engineering Organization

Phase	Duration	Main Initiatives	KPI
Phase 1: Data Foundation	0-12 months	• Installing IoT sensors (maintenance, quality) • Improving PLM/ERP BOM data quality • Digital thread phase 1 (PLM ↔ ERP integration) • AI pilot (image inspection and predictive maintenance on one line)	• Sensor coverage rate • BOM data quality score • Pilot defect rate/stoppage rate
Phase 2: AI Scale-Up	12-24 months	• Rolling out AI image inspection to all lines • Deploying SAP PM-integrated AI predictive maintenance to all equipment • Introducing Generative Design (integrated into the new-product-development process) • AI project management (SAP PS integration)	• Quality-cost reduction rate • Unplanned stoppage reduction rate • Design-lead-time reduction rate
Phase 3: The Intelligent Factory	24-36 months	• Completion of the digital-twin plant model • Real-time integration of ERP/MES/PLM/QMS • Autonomous production optimization by AI agents • Fully AI-driven project management	OEE improvement rate, time-to-market reduction rate, engineer productivity

Chapter 7: AI Use in EPC Projects and the Codification of Tacit Knowledge

Plant engineering and construction in the EPC (Engineering, Procurement, and Construction) domain is a mega-project industry involving enormous investment, extremely complex processes, and long-term uncertainty. Its success has depended on the advanced “tacit knowledge” of project managers (PMs), veteran designers, and construction engineers in process management and risk prediction. However, due to labor shortages and the aging of skilled workers, apprenticeship-style skill transfer has reached its limits, and the loss of know-how confined to individuals’ minds has become a common risk across the industry.

※ The cost of resolving trouble and rework in plant construction increases exponentially the later the phase in which it occurs. This is precisely why “front-loading”—incorporating construction risk during the design and initial planning stage—is critically important, and the codification of tacit knowledge through AI and large language models (LLMs) is the decisive tool for achieving it.

7-1 Digital Transformation in the EPC Domain and the Challenge of Tacit Knowledge

Japan’s three major engineering companies and global mega-contractors are rapidly advancing DX that visualizes and standardizes tacit knowledge into “explicit knowledge,” making it reproducible within systems. This chapter analyzes advanced cases at major EPC firms, general contractors, and overseas contractors, and systematically organizes the automation of process management, the prediction of design and construction risk, and the construction of digital knowledge platforms for skill transfer, together with their concrete mechanisms and results.

The analysis proceeds from three perspectives.

• Automation of process management: AI-driven schedule generation, delay prediction, and resource optimization

• Prediction of design and construction risk: front-loading of hazards using 3D CAD and sensing

• A knowledge platform for skill transfer: democratizing expert knowledge and permanently embedding it into systems via LLMs/RAG

7-2 Overview: A Three-Generation Model of AI Use and a Summary of Leading Case Studies

This chapter covers many case studies from Japan and overseas. As an overview, it first presents the developmental stages of AI use in EPC (a three-generation model), a comparison of the major leading cases, and the implications for Japanese companies. The individual case studies (from 7-3 onward) are best read as substantiating this framework.

A Three-Generation Model of AI Use and Implications for Japanese EPC Companies

Surveying the moves of leading overseas companies, AI use in EPC can be organized into the following three generations.

Generation	Positioning	Main Application	Representative Examples
1st Generation	Document/drawing creation support	Copilot-type design assistants	Various companies’ generative AI use
2nd Generation (current mainstream)	AI-based project controls	Progress/cost/risk prediction, automatic reporting	Fluor, Bechtel
3rd Generation (the next frontier)	Autonomous project management	Optimization of process/procurement, digital twins	HS2/ALICE, Bechtel x NVIDIA, nPlan

For Japanese plant EPC companies (JGC, Chiyoda Corporation, Toyo Engineering, etc.), the order of highest ROI can be considered as follows.

• Phase 1 (up to six months): AI process-delay prediction, AI procurement-delay prediction, automatic AI weekly-report generation (Fluor-type, i.e., learning from past projects)

• Phase 2 (1-2 years): AI cost-overrun prediction, AI risk-register generation, AI Lessons Learned search (nPlan-type)

• Phase 3 (3-5 years): digital-twin EPC, autonomous project controls, automatic AI process generation (Bechtel + NVIDIA + ALICE-type)

⚠ A shift in the competitive axis: leading overseas companies are no longer at the stage of “using generative AI to write minutes.” The competitive axis has shifted to “AI predicting future delays and cost overruns and getting ahead of project-control decision-making.” For Japanese EPC companies, which possess the asset of past project data, the Fluor-type model of learning from past projects is relatively easy to realize and can become a source of competitive advantage.

A Comparison of Case Studies of Major Leading Examples

The technology composition of the analyzed companies and systems, the tacit-knowledge/PM challenges they solve, and their main quantitative results are compared in the table below.

Company/AI System	Main Technology Composition	“Tacit Knowledge/PM” Challenge Solved	Main Benefits/Quantitative Results
Toyo Engineering AI for U	3D CAD risk-analysis AI, personnel data COMPANY/CTM 2.0	Predicting delay hazards in underground (UG) works, eliminating experience-dependent subjective judgment	10% reduction in contingency reserve, up to 90% reduction in plan-creation time
Chiyoda Corporation PlantStream	Autonomous 3D CAD, Unity engine, automatic PM defect flagging	Systematizing veteran designers’ spatial-routing experience, preventing oversights in design checks	About 75-80% reduction in initial spatial-design man-hours, 1,000 pipes routed automatically per minute; proposals shortened from weeks to days
JGC Group Project Digital Twin	Reinforcement learning/Generative Design, AWP, ISO 8000 data management	Optimal plot-plan design, dynamic management of the gap between plan and actual	Automation of routine checks and shortened design time, maximizing AI prediction accuracy through advanced data management
Shimizu Corporation Lightblue Assistant	RAG (image-processing optimization, chunking, Teams integration), motion-sensor AI	Referencing a 1,000+ page technical manual, personalization of on-site handovers, physical inspection judgment for rebar	RAG accuracy 35% → 93%, search time 30 minutes → a few minutes, 12,000 accesses in 2 months, gas-pressure weld inspection 5 minutes → 20-30 seconds
Obayashi Corporation Blast Skill AI	Ground-assessment AI engine, Salesforce workflow standardization	Skilled blast design tailored to geology, personalization of approval know-how for non-routine work	Automatic creation of blast patterns linked to ground conditions, standardized business procedures; freeing junior staff for creative work
Bechtel Proprietary LLM/Predictive AI	Proprietary LLM, procurement-risk AI, digital twin	The enormous effort of checking O&M manuals, supply-chain delivery delays	Manual comprehension reduced from days to minutes, 20% reduction in supply-chain delays
ALICE Technologies AI Scheduling	Generative scheduling AI, P6/BIM import integration	Redesigning the critical path under experience-dependent resource constraints	17% shorter construction period, 14% lower labor cost, 12% lower equipment/crane operating cost

7-3 AI Implementation at Major Japanese Engineering Companies

Toyo Engineering: Constructability Review and Personnel/Risk Management with “AI for U”

Toyo Engineering is pursuing a company-wide DX strategy, “DXoT (Digital Transformation of TOYO),” aimed at a substantial increase in project productivity. Developed in partnership with the AI startup HEROZ, its constructability-review system “AI for U” is the result. In the industry it is said that “whoever masters underground (UG) work masters construction”—excavation, drainage, and other underground work is highly susceptible to delay risk from weather and ground conditions. Conventionally, foreseeing such hazards relied on veteran engineers’ subjective, localized judgment (tacit knowledge), which was difficult to accumulate systematically or pass on to younger staff.

• Mechanism: automatically detects delay hazards during underground construction from a 3D CAD model and automatically pushes quantitative grounds for their occurrence at the design stage. Designers can build the risk into the drawings in advance, preventing sudden rework and delays at the construction stage (development period of about 1.5 years).

• A dual effect: for veterans, it provides “the reassurance of having intuition validated by data”; for junior staff, it functions as an educational tool for autonomously learning “why this layout is dangerous.”

• Digitizing personnel/PM knowledge: by introducing WHI’s “COMPANY” and the talent-management system “CTM 2.0,” skills, project experience, and career aspirations are managed centrally. Staffing decisions that had depended on a PM’s personal network and subjective judgment were transformed into agile, optimal assignments made with an eye toward development.

[KGI/Results] The company anticipates building a model that reduces the contingency reserve covering estimate uncertainty by 10% and cuts plan-creation lead time by up to 90%. Company-wide, the goal is a substantial increase in project productivity.

Chiyoda Corporation: Autonomous Spatial Design and the Democratization of Tacit Knowledge with “PlantStream”

Chiyoda Corporation has set out data-centric EPC reform under its mid-term management plan “Management Plan 2025,” and its greatest achievement is the autonomous 3D CAD “PlantStream,” developed as a joint venture with the CAD-development startup Arent. Large-scale oil, gas, chemical, and hydrogen plants contain thousands to tens of thousands of pipes, and “spatial routing design”—safely and efficiently connecting their start and end points—was labor-intensive, personalized work in which veterans spent hours calculating pressure loss and avoiding interference in their heads.

• Technology composition: the Unity game engine’s lightweight 3D rendering power is fused with the company’s vast design know-how encoded in a routing algorithm (logical design), achieving the ability to automatically lay out as many as 1,000 highly precise pipes per minute.

• The philosophy of “democratizing tacit knowledge”: Arent CEO Hiroki Kamobayashi positions the essence of the effort as “democratizing tacit knowledge and globally commercializing it as a SaaS.” The company aims for a “platformization of knowledge” reminiscent of how Dassault Systèmes spun off its CAD division to grow it into a global company.

• Automated checking: the PM system automatically extracts defect data and customer-facing points of caution from past similar design changes, preventing oversights and omissions in design checks.

[External sales case] A leading U.S. EPC contractor, S&B, adopted PlantStream. Presenting multiple design options to a client, which had previously taken several weeks, was completed in just a few days, achieving a significant reduction in initial design time and improved MTO (material take-off/estimate) accuracy.

※ In addition to subsidizing G-Kentei and E-Shikaku certifications, the company has established a “DX competency” personnel-evaluation system that is directly reflected in salary, fostering a company-wide culture of using digital technology effectively. That the effort is backed not only by technology but also by institutional policy is instructive.

JGC Group: The IT Grand Plan 2030 and the Project Digital Twin

The JGC Group has formulated its “IT Grand Plan 2030,” pursuing DX across the entire EPC lifecycle. Rather than stopping at automating individual processes, it aims to optimize project operations as a whole.

• Automated design AI: reinforcement learning and generative design perform optimal placement and automatic selection of plant equipment. Senior-level checking work has been automated, evolving from personalized, single-option design to instant AI-driven creation and comparison of multiple options.

• Project digital twin: progress, budget, and risk factors are synchronized in real time within a computational model for simulation, predicting future schedule delays to support decision-making. Advanced management methods such as AWP (Advanced Work Packaging), which dynamically visualizes and controls the gap between plan and actual, are incorporated.

• Ensuring data quality: with support from Hitachi, the company has established an advanced data-management process compliant with the international data-quality standard ISO 8000. It incorporates PDCA/OODA loops that ensure completeness, consistency, and accuracy, raising the accuracy of AI output and the reliability of predictive management.

7-4 AI Use at General Contractors and the Codification of Skilled Expertise

Shimizu Corporation: Advancing a Knowledge Platform with Generative AI x RAG

As a countermeasure to on-site labor shortages, Shimizu Corporation built a skills-transfer platform using generative AI and RAG (retrieval-augmented generation). It deployed Lightblue Inc.’s “Lightblue Assistant” on Microsoft Teams to launch a company-wide knowledge-sharing system. The greatest challenge was accurately feeding a specialized reference work of more than 1,000 pages across three volumes, “Fundamentals of Building Construction,” into the AI so that it could answer questions from the field.

• Accuracy improvement: at first, RAG could not properly handle specialized drawings, formulas, and complex comparison tables, and response accuracy languished at around 35%. Through technical improvements such as optimized image processing, page-level chunking, and Teams integration, response accuracy was dramatically improved to 93%.

• Reduced search time: searches for construction procedures, building codes, and precautions for concrete placement, which previously took more than 30 minutes, were dramatically shortened to a few minutes.

• The “My Assistant” feature: site personnel can create their own assistants, and roughly 50 individual-purpose assistants have emerged. Examples include a “summary master” that extracts only the decided items from lengthy documents, and a “handover reference” trained on past site managers’ handover documents—capturing the hands-on tacit knowledge unique to the field.

• Sensing skilled techniques: the work movements of skilled craftsmen are captured with sensors and cameras, and the “gas pressure welding joint inspection” of rebar—difficult to put into words—is judged by AI (inspection time reduced from 5 minutes to 20-30 seconds).

[Proof-of-concept results] within two months of the trial rollout, more than 12,000 accesses were recorded across 18 sites and 260 accounts, of which 46% (about 5,500) were RAG searches of manuals and construction drawings, achieving an extremely high utilization rate.

Obayashi Corporation and Skanska: Blast Skill AI and Safety Sidekick

• Obayashi Corporation’s “Blast Skill AI”: in mountain tunnel construction, this AI-enables the tacit knowledge of “blast design”—veterans assessing the geology and determining the placement and strength of dynamite. By training the model on the ground-assessment judgment process, it automatically creates the optimal blast pattern matched to the hardness and shape of the ground.

• Obayashi’s business standardization: day-to-day decision-making, business applications, and approval procedures are built into standardized patterns on Salesforce, standardizing work methods that had previously stayed with individuals and freeing junior staff to focus on creative management work without hesitation.

• Skanska’s “Safety Sidekick”: the company’s safety-management manual, EHS standards, and the U.S. OSHA standard documents are fed into a GPT-4o-based AI, allowing workers to check safety measures for a work scenario via voice or text from their smartphone. As veteran retirements progress, it serves as an effective means of embedding safety education backed by real-world experience for new employees.

7-5 AI Governance and Generative Scheduling at Overseas Mega-Contractors

Bechtel (USA): A Data-Intensive AI Assistant

In February 2026, Bechtel appointed John Platt as Chief Technology Officer, driving a large-scale overhaul of its “project delivery model” that fuses AI, automation, robotics, and 3D printing. The company uses AI to decipher the correlations within the enormous volumes of data generated at infrastructure, semiconductor-fab, and large-scale data-center construction sites, optimizing the value chain.

• AI assistant: thousands of pieces of precision equipment each come with thousands of pages of O&M (operations and maintenance) manuals. The company structures these tens of thousands of pages of technical documentation and feeds them into its own proprietary LLM, so that the necessary protocols, key points, and task lists can be generated instantly from a natural-language query.

• Testimony from the field: functional department manager David Wilson states that “work that used to take humans days to review O&M requirements has been shortened to minutes, letting people focus on more difficult, essential work.”

[Results] predictive AI implemented in procurement and the supply chain reduced delivery-delay risk by 20%. Significant results have also been achieved in process changes for on-site construction.

ALICE Technologies x McKinsey: Generative Scheduling

Among AI applications in EPC project management, “generative scheduling” is where AI delivers the most revolutionary results. The U.S.-based ALICE Technologies leads this area and has formed a strategic alliance with McKinsey & Company to transform capital projects worldwide. With conventional tools such as Primavera P6, planners had no choice but to assemble resources and work sequences one-dimensionally based on experience (tacit knowledge), and rebuilding the schedule during unexpected delays or resource crunches tended to lag.

• Mechanism: ingesting BIM models or P6 data, it parameterizes “labor,” “construction equipment,” “materials,” “space (interference),” and “work-sequence dependencies,” autonomously generating millions of construction-schedule patterns (what-if scenarios).

• Dynamic optimization: the schedule is automatically re-synced (Schedule Sync) in response to progress changes, executing critical-path analysis and automatic logic-loop detection in real time. It helps select the construction path with the highest profit margin and lowest uncertainty.

[Results] an average project-wide reduction of 17% in construction duration, 14% in on-site labor cost, and 12% in equipment/crane operating cost. The largest cost compressions have been reported on projects run jointly with McKinsey.

7-6 Leading Overseas EPC Case Studies: The Implementation Phase of AI-Based Project Controls

The previous sections mainly looked at design, construction, and the codification of tacit knowledge. This section organizes, to the extent supported by publicly available information, cases in which leading overseas EPC companies are applying AI to “project controls itself” (schedule, cost, procurement, and risk management). What they have in common is that rather than the AI reporting “the current state,” it predicts “what will happen in the future (delays, cost overruns)” and gets ahead of decision-making.

Fluor (USA): Predictive Project Health Diagnostics (IBM Watson)

In 2018, Fluor worked with IBM Watson to develop an AI system that predicts, monitors, and measures the state of an EPC mega-project from proposal through completion. Its core consists of two components: “EPC Project Health Diagnostics (EPHD)” and “Market Dynamics / Spend Analytics (MD/SA).”

• Data integration: fuses thousands of data points spanning design, procurement, construction, fabrication, and the supply chain, drawing out dependencies and practical insights

• Prediction: predicts problems such as cost increases and schedule delays from past trends and patterns, gaining early insight from a complex web of factors

• Domain-specific UI: a natural-language, conversational interface specialized for the EPC field provides access to data, reports, and results

[Takeaway] what Fluor uses is not so much a novel technology as “learning from past project data” (schedule, cost, changes, and procurement history). For Japanese plant EPC companies that hold past-project databases and Lessons Learned/Change Order histories as an asset, this is the most readily imitable approach.

Bechtel (USA): AI-Based Project Controls and the Digital Twin (NVIDIA Omniverse)

Aiming to improve productivity across its EPC business, Bechtel has established a structure in which Senior Vice President John Platt leads its EPC transformation. In addition to the O&M assistant described in the previous section, the company has also embedded AI directly into project-control operations.

• AI project controls: automates schedule updates, performance reports, and progress analysis, shortening work that previously took several days. The flow is shifting toward “project data → AI analysis → management decision.”

• Digital twins with NVIDIA Omniverse: in the construction of AI factories and similar facilities, design, procurement, construction, and commissioning are turned into a digital twin. Before construction begins, heat, power, and layout are simulated in a physically accurate 3D environment to perform failure prediction and optimization.

• Modularization: standardized components combined with integrated design, procurement, construction, and commissioning shorten the time to start-up.

※ Bechtel’s direction is “Project Data → AI Analysis → Management Decision.” The gap with Japan’s multi-step process of “Project Controls department → analysis → report writing → management meeting” points to an issue Japan’s EPC industry will need to confront over the next five years.

HS2 (UK): Implementing Generative AI Scheduling (ALICE x SCS JV)

On the UK’s HS2 high-speed rail project, a joint venture of Skanska, Costain, and STRABAG (SCS JV) implemented the generative AI process-optimization (“optioneering”) platform ALICE Technologies. For the construction of a roughly 13-mile twin-bore tunnel on the London section, the AI explored a vast number of construction scenarios to optimize work sequencing, crew placement, and equipment allocation.

• Quantitative result 1: through what-if analysis, identified a way to shorten the construction period of the Euston shaft by 86 working days

• Quantitative result 2: achieved a cost saving of approximately £2 million (2 million pounds) for the target scope

• Quantitative result 3: created more than three months of float (schedule slack) in the Copthall Green tunnel process

[Takeaway] process management is shifting from “creation and revision through experience” to “exploration through an AI algorithm.” Rather than an extension of Primavera or MS Project, this is a symbolic case showing that AI has entered the stage of “thinking through” the schedule itself.

nPlan: Predicting Schedule Risk by Learning from Past Actuals (Suffolk/LNG Canada)

nPlan is an AI that trains a deep-learning model on the world’s largest database of “planned plus actual” schedules (more than 750,000 program files), predicting the actual outcome of every activity in any given schedule.

• Mechanism: simply by reading in a schedule, it predicts processes prone to delay and hidden risks, modeling “how many days could be saved if the risk is properly addressed”

• Case (Suffolk Construction): major U.S. general contractor Suffolk used nPlan as its AI risk-management and prediction partner on the expansion of a prominent Boston hospital (an 11-story, over 550,000-square-foot complex project including a NICU and hybrid operating rooms)

• Case (LNG Canada): on one of the world’s largest LNG projects, used for risk prediction, schedule prediction, and advancing stage-gate decision-making

※ nPlan’s value lies not in showing “what the current state is” but “what the future state will be.” Learning from a large volume of past actuals surfaces “hidden delay risk” that does not appear on the critical path, before construction even begins.

7-7 Visualizing Tacit Knowledge and Systematically Integrating AI Use (the SECI Model)

These cases can be clearly positioned by mapping AI technology onto the “SECI model” of organizational learning (the mutual conversion of tacit and explicit knowledge). AI elevates knowledge that had been confined to a veteran individual’s experience into an organizational, global asset.

Conversion Mode	Flow of Knowledge	Conventional Means	Example of AI Application in EPC
Socialization	Tacit → Tacit	Sharing experience through OJT and apprenticeship	Sensing skilled workers’ movements with sensors/cameras (Shimizu: gas-pressure welding; Obayashi: ground assessment)
Externalization	Tacit → Explicit	Putting experience into language and formulas	Decomposing judgment into parameters and modeling it with AI (AI for U, Blast Skill AI)
Combination	Explicit → Explicit	Systematizing documents and drawings	Integrated search of manuals and drawings via LLM/RAG (Bechtel, Shimizu’s Lightblue)
Internalization	Explicit → Tacit	Embodying explicit knowledge through practice	Junior staff learning “why this is dangerous” through the AI’s stated rationale (AI for U as an educational tool)

※ AI does not merely automate tasks; it accelerates all four modes of the SECI cycle. The essential value—particularly in permanently embedding “externalization” (tacit → explicit knowledge) as a system—lies in converting the crisis of skill transfer into an organizational asset.

7-8 Conclusion: The EPC PM Transformation Brought by AI and Organizational Challenges

These leading cases point to common success factors for establishing AI in practical use in the EPC and construction domain, as well as the governance challenges that must be overcome. The key points can be summarized into three.

• (1) The “quality” and “management structure” of input data: as with the JGC Group’s introduction of ISO 8000-compliant data-quality management, the accuracy of AI predictions is entirely proportional to the degree to which the underlying data has been organized (garbage in, garbage out). Integrating documents, drawings, and ERP/PM data across departments onto a common, data-centric platform is essential infrastructure for maturing AI into practical use.

• (2) Bridging domain knowledge and AI (bridging talent): in developing AI for U, intuitive tacit knowledge such as “deep excavation is prone to collapse in wet weather” was decomposed into mathematical parameters such as soil strength, excavation angle, forecast rainfall, and the load of surrounding piping. People who understand both on-site practice and AI algorithms are essential, and Chiyoda Corporation’s system of tying DX qualifications directly to salary evaluation is an advanced benchmark for an organizational solution.

• (3) Clarifying the division of responsibility between humans and AI (AI governance): the principle that “AI is a support tool, and the final decision is made by a human” must be thoroughly upheld. When handling mega-project technical documents that are a treasure trove of confidential information, governance against security risk and hallucination is essential—including risk classification before use, establishing masking rules, and clarifying accountability.

⚠ In response to the crisis of skill transfer, the democratization of tacit knowledge through AI and its permanent embedding into systems (as with PlantStream and AI for U) elevates individually held know-how into a “reproducible, scalable organizational asset (an explicit-knowledge platform).” A process in which AI autonomously generates millions of plans and humans review, evaluate, and collaborate on them will become the source of competitive advantage in capital-intensive industries going forward.

Summary of AI Impact Across Engineering Domains

Domain	Main AI Technology	Representative Effect (Measured Basis)	Realization Timeframe
Design & Development	Generative Design, AI surrogate CAE, automated PLM-ERP BOM	30-60% shorter design time, 30-60% weight reduction, 80% shorter design-change reflection time	12-24 months
Quality (Image Inspection)	Computer vision (CNN), AI-enabled AOI	35% reduction in defect rate, 60-80% reduction in inspection cost, stable 24-hour detection	3-12 months
Quality (SPC)	Multivariate AI-SPC, root-cause analysis AI	50-60% reduction in process defect rate (leading cases)	6-18 months
Equipment Maintenance	AI predictive maintenance, IoT x SAP PM integration	20-40% reduction in unplanned stoppages, 25-40% reduction in maintenance cost, 15-30% reduction in spare-parts inventory	6-18 months
Project Management	AI-EVM, SAP PS integration, delay prediction, resource optimization	20-30 point improvement in on-time project completion rate, ±3-5% cost-prediction accuracy achieved	9-18 months
Software Development	AI coding assistants (Copilot, etc.)	30-55% increase in development speed, automatic test generation, reduced code-review effort	3-6 months (tool rollout)

(This report was prepared based on Engineering.com, Siemens, Autodesk, PwC, and various manufacturing implementation case studies (2025-2026). The reported figures are typical values across multiple implementation cases and will vary by individual environment.)