Industrial Digital Twin & AI Application Solution Engineer
I focus on the connection between industrial systems, digital twins and AI applications. My studies and work have introduced me to production processes, automation, simulation, human-system interaction and software tools. They have also led me to ask how scattered engineering information can be organized so complex systems become easier to understand, verify and use. I now want to keep exploring how AI can enter real workflows and help people get work done.
Current
AI Solutions Specialist
Education
TUM M.Sc. · Tongji B.Eng.
Focus
AI Applications · Industrial Simulation · Digital Twins
Production systems, process analysis, operations research and industrial engineering methods, building an understanding of manufacturing through complete processes and resource relationships.
2021–2024
Technical University of Munich · Munich
M.Sc. Human Factors Engineering
Human-system interaction, interface design, data analysis and engineering systems, with a focus on how technology is understood, used and accepted.
2023–2026
BMW Group · Munich
Digital Twin / Virtual Commissioning
Industrial simulation, virtual commissioning, digital training and system integration development, with practical experience of how digital models enter real engineering workflows and create value.
CURRENT DIRECTION
AI Applications & Digital Simulation
Connecting simulation tools, industrial data and explainable intelligent systems.
LIVE / ASM-A01THREE.JS · INDUSTRIAL SCENE SIMULATION● Online
Starting industrial scenePreparing the lightweight mobile renderer…
02 / SELECTED WORK
Explore three engaging scenarios
Explore engaging projects from enterprise practice, personal exploration and research, and experience different ideas across industrial simulation, tool building, data analysis and human-system interaction.
BMW Group · Enterprise SystemTechnical Project
Industrial Fault Simulation & Training System
Extend virtual-commissioning assets into a digital-twin training platform with automated station recognition and standardized fault inheritance.
OutcomeThe global rollout supported 1000+ training users, standardized 200+ reusable fault scenarios and improved troubleshooting capability by 25%.
C#
.NET
PLCSim
OPC UA
Fault Injection
Operation steps
01Station scan & readinessChoose a role and station, identify equipment, signals and training objects, then verify the sandbox.
02Standard fault inheritanceGenerate station scenarios from a standardized fault catalog, with full CRUD controls.
03Sim & verifyOne fault drives PLC signals, HMI alarms and the 2D model together.
04Training reportValidate diagnosis and recovery answers and retain the execution trace.
Try it yourselfExplore the project in action
Interactive demoPrepares automatically as you approach
Project context
Project summary
A digital-twin training platform that reuses models and control logic created for virtual commissioning, then extends them into reproducible fault training and verification throughout the production-system lifecycle.
200+reusable fault scenarios
1000+training users supported
25%troubleshooting improvement
Problem / overview
Industrial troubleshooting training depended on real equipment and non-reproducible failures, making it expensive, risky and difficult to scale. Building a separate digital twin would add substantial modeling cost, while models already created for virtual commissioning were often underused after launch. The opportunity was to extend those assets across the production-system lifecycle, reduce duplicate engineering effort and establish a new training service.
Solution
The platform connects the virtual-commissioning model, PLCSim and the OPC UA-based HMI through a shared interface layer. It identifies station equipment, signal points and training objects, inherits applicable scenarios from a standardized fault catalog, and automatically generates the station-level training configuration. Multi-layer fault injection then synchronizes PLC variables, digital-twin object behavior and HMI alarms, allowing users to configure, reproduce, diagnose and verify the complete fault-and-recovery workflow without disrupting production equipment.
Result / impact
The platform was deployed across five factories in Germany, Hungary, Mexico, China and the United States, including development and training use cases for next-generation battery production. The global rollout supported 1000+ training users, standardized 200+ reusable fault scenarios and improved troubleshooting capability by 25%.
Highlights
Identified station equipment, signal points and training objects to reduce manual modeling and configuration work.
Inherited applicable scenarios from a standardized fault catalog to generate station-level fault libraries and training configurations.
Injected faults across signal level, simulation object level, and HMI interaction level.
Used an interface layer to bridge simulation runtime, PLCSim, and OPC UA-based HMI control.
Built scenario-based fault logic so training could move from passive troubleshooting to deliberate validation.
Technologies
C#
.NET
PLCSim
OPC UA
Fault Injection
Simulation
HMI
System topology
Recognized station content inherits the standardized fault catalog to form a station training configuration before three verification targets are driven in sync.
01Station recognitionEquipment, signal points and training objects
02Standard fault inheritanceStation scenarios and training configuration
03Synchronized interfaceUnified state, signal and command mapping
PLC / PLCSimInputs, outputs and safety interlocksSimulation ModelObject behavior and visual responseHMI / OPC UAAlarms, state and operator feedback
S.Hu · Personal ProjectExtended Technical Project
Industrial Simulation Data Agent
Use structured documents to describe industrial scenes, interpret configuration intent, generate digital twins, and support defect simulation and engineering outputs in one framework.
OutcomeStructured YAML and 3D transforms stay synchronized across four reproducible defect cases, with USD, metadata and camera outputs generated from one scene state.
NVIDIA Isaac Sim
OpenUSD
YAML
Synthetic Data Generation
Industrial Digital Twin
Interaction path
01Interpret & configureMap scene requirements to structured documents, asset references and parameter rules.
02Generate the digital twinAssemble industrial components, relationships and the USD scene tree from the document.
03Simulate defects & errorsReuse one scene definition across normal operation and four parameterized anomalies.
04Compose outputs on demandGenerate hierarchy trees, reports, metadata and scene captures from one scene state.
Try it yourselfExplore the project in action
Interactive demoPrepares automatically as you approach
Project context
Project summary
A configuration-driven industrial simulation framework that turns structured scene descriptions into digital twins, while supporting defect simulation, 3D parameter write-back and on-demand outputs.
YAML ↔ 3Dbidirectional scene mapping
04reproducible defect cases
USD + JSONshared output state
Problem / overview
Industrial digital twins and simulation-data production still rely heavily on manually building USD scenes, placing assets and tuning parameters. Each scene variant or defect case often requires repeated modeling and a new set of output files, increasing engineering cost, weakening reproducibility and making scene knowledge difficult to reuse across stations and projects.
Solution
The framework connects intent interpretation, structured scene description, digital-twin construction, defect simulation and output generation in one configuration-driven workflow. Scene requirements are converted into YAML, which assembles reusable assets, materials and object relationships into a USD Stage and hierarchy. Direct 3D edits write transforms back to the document, normal and parameterized defect states share the same scene definition, and reports, hierarchy trees, metadata and scene captures are generated on demand from the synchronized state.
Result / impact
The framework combines scene construction, defect reproduction and engineering outputs into one reusable workflow. A single structured document can continuously drive the digital twin and multiple output formats, while assets, intent rules, defect models and output modules remain independently extensible for use across stations, projects and simulation-data tasks.
ISAAC SIM APPLICATION LOOP
Highlights
Uses structured documents as the shared entry point for scene configuration and 3D construction.
Translates scene intent into asset, parameter and relationship configurations.
Builds digital twins from reusable industrial components and a modular USD asset library.
Represents defects as parameterized states so normal and abnormal conditions share one scene definition.
Synchronizes YAML, 3D transforms and the USD hierarchy around one scene state.
Extends assets, defect rules and output modules independently, including hierarchy trees, reports and scene captures.
Technologies
NVIDIA Isaac Sim
OpenUSD
YAML
Synthetic Data Generation
Industrial Digital Twin
AI Agent
Scene Configuration
USD Asset Composition
Configuration-driven application loop
Scene requirements are interpreted into structured documents that automatically generate the digital-twin scene and hierarchy. Direct 3D edits and defect simulation keep one scene state current, while reports, hierarchy trees and scene captures are produced from it on demand.
01Scene intentInterpret task requirements into configuration goals and rules
02Structured descriptionYAML defines assets, parameters, relations and defects
03Digital-twin runtimeBuild the USD Stage and maintain one shared state
04On-demand outputsGenerate engineering evidence from the scene tree
Document ↔ 3DBidirectional structure and transform synchronizationDefect & error simulationNormal and parameterized abnormal states share one sceneHierarchy & visual evidenceScene trees, reports, metadata and captures
TUM · Human Factors ProjectMotion Evaluation Research
Vision-based Motion Evaluation & Feedback System
Turn abstract Qigong movement into personalized scores, statistical evidence and joint-level feedback, validated with more than 20 participants.
OutcomeThe study validated relative scoring across 6 joint groups with 20+ participants and integrated dynamic and cohort statistics into a reusable motion-evaluation framework.
Human Factors
Experimental Design
MATLAB
RStudio
Statistical Analysis
Interaction path
01Compare performanceSwitch between a reference movement and a sample that needs correction.
02Test a local correctionLocate a joint and drag it to explore a possible correction direction.
03Inspect the evaluation basisCompare participant-relative and absolute scoring across five kinematic metrics.
04Validate cohort evidenceChange sample size and move between box plot, central tendency and precision views.
Try it yourselfExplore the project in action
Interactive demoPrepares automatically as you approach
Project context
Project summary
A human-factors study that turns expert movement into a reusable relative-scoring framework, combining experiment design, kinematic data processing, statistical visualization and participant-specific feedback.
20+Participants
06Joint groups
05Kinematic metrics
Problem / overview
Qigong quality is abstract and difficult to evaluate with one absolute target. Body proportions, mobility and movement amplitude differ between participants, while visual coaching provides little objective evidence about which joint or phase needs correction. The study therefore needed a measurable but individualized evaluation basis, a statistically defensible comparison across participants, and evidence for how teaching videos should frame the movement.
Solution
The study defined the experiment, reference actions and scoring logic from experienced athletes and teaching videos. A company partner supplied the computer-vision skeleton sequences; based on the movement characteristics of Qigong, the project researched and selected joint angle, velocity, acceleration, angular velocity and angular acceleration as the evaluation dimensions, then cleaned and processed the sequences in MATLAB and RStudio. Correlations and movement patterns were used to build participant-relative scores rather than relying on raw magnitude. Dynamic joint diagnostics were combined with box plots, mean, median, quartiles and confidence intervals so individual feedback and cohort evidence could be examined together. Eye tracking added areas-of-interest evidence and practical filming guidance for teaching videos.
Result / impact
Testing with more than 20 participants connected participant-relative joint scoring with dynamic and cohort-level evidence. Statistical views exposed variation, central tendency and uncertainty, while larger participant samples made the evaluation basis progressively more stable. The outcome was a reusable experimental and data-evaluation framework for other sports, plus evidence-based filming guidance from eye-tracking areas of interest.
Highlights
Defined the experiment and objective scoring criteria for an otherwise abstract movement from expert athletes and teaching videos.
Kept the project boundary explicit: the company partner supplied CV skeleton sequences; study design, data processing, scoring and evaluation formed the project contribution.
Used relative correlation across angle, velocity, acceleration, angular velocity and angular acceleration instead of one absolute threshold.
Connected dynamic neck, shoulder, elbow, wrist, hip and knee feedback with cohort-level statistical analysis.
Used box plots, mean, median, quartiles and confidence intervals to reveal variation and show greater evaluation stability as the participant sample grows.
Validated the framework with more than 20 participants and used the feedback to improve Qigong performance.
Combined eye-tracking areas of interest with motion analysis to define teaching-video filming guidance.
Designed the data and visualization pipeline as a reusable evaluation framework for other sports.
Technologies
Human Factors
Experimental Design
MATLAB
RStudio
Statistical Analysis
Data Visualization
Relative Scoring
Skeleton Data
Eye Tracking
Experiment, data processing and evaluation loop
A company partner supplied CV skeleton sequences as the data input. The project contribution begins with experiment and scoring-criteria design, then spans kinematic processing, relative evaluation, statistical validation and visual feedback, with eye tracking adding teaching-video guidance.
01Experiment & criteriaDefine the evaluation basis from experts, teaching videos and research questions
02Metric research & processingSelect five metrics from the movement characteristics, then process six joint groups
03Relative scoring & statisticsCombine individual movement patterns, cohort distributions and uncertainty
04Dynamic & cohort visualizationReturn joint feedback, box plots, central tendency and precision
Individual & cohort evidenceConnect joint feedback, statistical distributions and evaluation stabilityReusable evaluation frameworkTransfer experiment, data, scoring and visualization to other sportsTeaching-video guidanceTurn eye-tracking areas of interest into filming rules
03 / OTHER PROJECTS
More projects and practical work
More enterprise projects, personal work and research practice. Expand an item to read its complete background, solution, result and technologies.
Personal Project
Production Line Auto Builder
A Unity-based tool that rapidly generates industrial production-line scenes from structured JSON, with component inspection and basic fault simulation.
+
Project summary
A Unity-based tool that rapidly generates industrial production-line scenes from structured JSON, with component inspection and basic fault simulation.
Problem / overview
Manually assembling industrial scenes and organizing their components can be time-consuming, making it difficult to quickly validate different production-line structures. This project explores a structured approach to scene generation while keeping component identities and functions clear and traceable.
Solution
The system reads JSON configuration to automatically generate a Line / Station / Device hierarchy in Unity and names components according to consistent rules. The same data structure can produce simple or complex scenes while preserving layout, component relationships, and functional information.
Result / impact
The resulting prototype validates a configuration-to-scene workflow for rapid production-line modeling and interactive representation. It focuses on fast modeling and system communication rather than full industrial simulation fidelity.
Project evidence · Production Line Auto Builder
Key features
JSON-driven production-line modeling.
Automatic Line / Station / Device hierarchy generation.
Conveyor, Sensor, RobotBase, and other component types.
Consistent structured naming for generated components.
Simple and complex scene layouts.
Conveyor blockage triggering and recovery.
Interaction
Navigate the scene using WASD and the mouse.
Hover over components to inspect information such as name, type, and function.
Switch between simple and complex layouts.
Trigger and reset a conveyor blockage.
Technologies
Unity
JSON
Configuration-driven Design
Scene Interaction
Fault Simulation
Personal Project
Aster Desktop Assistant
A personal desktop assistant that keeps track of long- and short-term goals, follows topics of interest, and delivers relevant news and information.
+
Project summary
A personal desktop assistant that keeps track of long- and short-term goals, follows topics of interest, and delivers relevant news and information.
Problem / overview
In the fast-moving AI era, useful information is easily buried in constant updates. People often need to repeatedly search for news, remember unfinished plans, and reconnect scattered information with their personal goals. Aster brings these needs together in a desktop companion that stays close at hand.
Solution
Instead of waiting for the user to repeatedly search and reorganize information, Aster connects goals, reminders, interests, and news in one continuous personal context. It helps turn incoming information into something relevant, timely, and easier to act on.
Result / impact
Aster brings conversations, goals, reminders, and selected news into one desktop experience, reducing the effort needed to follow important information and keep daily actions connected to longer-term direction.
Project evidence · Aster Desktop Assistant
Key features
Keep long-term goals visible and easy to revisit.
Turn short-term plans into reminders and next actions.
Follow personal interests and preferred topics.
Collect and filter relevant AI news and updates.
Deliver personalized information directly on the desktop.
Support natural text and voice conversations.
Remember useful context across conversations.
Technical highlights
Keeps personal goals, interests, and conversation context available across sessions.
Stores personal information locally for a more private experience.
Accepts voice input even when online speech services are unavailable.
Displays responses as they are generated for a smoother conversation.
Lives directly on the desktop instead of requiring a browser tab.
Interaction
Hover to quickly preview the assistant.
Click to open conversations, goals, reminders, and news.
Drag Aster to a comfortable position on the desktop.
Use voice or text to record plans and interests.
Receive reminders and relevant updates through desktop notifications.
Technologies
Electron
React
TypeScript
SQLite
OpenAI API
Whisper
Desktop Agent
Personal Project
PLC Selection Decision Assistant
An AI-assisted decision workspace that helps engineers compare PLC platforms by combining objective evidence with project experience and preferences.
+
Project summary
An AI-assisted decision workspace that helps engineers compare PLC platforms by combining objective evidence with project experience and preferences.
Problem / overview
Equipment selection in engineering rarely depends on a single specification. Engineers must balance performance, cost, existing systems, team experience, future maintenance, and project risk. This project explores how AI can support that decision process by organizing requirements, comparing candidate PLC platforms, and explaining the trade-offs between objective results and practical preferences.
Solution
The decision method separates evidence from judgment without forcing one to replace the other. Objective comparison provides a stable baseline, project priorities reflect real engineering conditions, and AI connects both sides by explaining trade-offs, questioning incomplete assumptions, and organizing the reasoning into a reviewable recommendation.
Result / impact
The assistant turns an experience-heavy selection discussion into a transparent decision record. Engineers can see why a platform ranks higher, where personal preferences influence the result, which information is still missing, and how the final recommendation was formed.
Create a selection project, enter its goals, scale, existing platform, constraints, and team experience, then add reference materials and candidate PLC platforms.
Set the comparison criteria and project priorities, then run the platform analysis.
Review the objective comparison together with AI explanations, risks, missing information, and recommended next steps.
Adjust the decision when needed, edit the generated report, and export the final selection recommendation.
An industrial AI assistant that helps interpret production-line conditions, route diagnostic tasks, and explain likely fault causes.
+
Project summary
A technical exploration of how an Agent can combine production status, alarms, operating context, and specialized tools to support repeatable and explainable fault diagnosis in industrial environments.
Problem / overview
Fault diagnosis in industrial environments requires engineers to connect alarms, equipment status, production context, and past observations under time pressure. AI can help organize these clues, route each question to the right source, compare possible causes, and keep the diagnostic path visible instead of reducing the task to a one-off answer.
Solution
The solution explores an Agent workflow that plans the diagnosis, calls the appropriate industrial tools, reviews the collected evidence, and revises the route when information is incomplete. Memory keeps multi-turn context connected, while a visible execution record allows each conclusion and recommendation to be reviewed.
Result / impact
The project shows how an industrial Agent can move beyond conversation and support a complete diagnostic path: organizing evidence, routing tools, comparing causes, preserving context, and making every recommendation reviewable. It provides a practical foundation for exploring explainable AI assistance in manufacturing fault analysis.
Project evidence · Production Line Status & Fault Diagnosis Agent
Key features
Understand fault descriptions together with current production-line conditions.
Collect equipment status, alarms, and relevant operating information.
Send each diagnostic question to the most suitable information source or tool.
Compare symptoms with possible fault causes and supporting evidence.
Keep previous questions and observations connected during the diagnosis.
Present likely causes, supporting clues, and recommended next actions.
Technical highlights
Plan–Execute–Review/Replan workflow replaces one-shot model responses with staged diagnosis.
Planner, Executor, Reviewer, and Orchestrator modules keep responsibilities separated.
Function calling routes diagnostic tasks across multiple industrial tools.
Structured semantic interfaces unify system state, alarms, and production context.
Short-term, summary, and structured memory preserve continuity across diagnostic turns.
Context continuation and coreference handling support follow-up questions without losing the fault context.
Synchronize PLC-controlled equipment states with Unity through OPC UA and REST, presenting the industrial control process as a complete digital chain that reproduces a production digital-twin workflow.
+
Project summary
A compact industrial digital-twin system that connects PLC logic simulation, industrial communication, application services, and a Unity scene so equipment and logic state changes can be observed and understood across the complete data path.
Problem / overview
Control logic, communication services, and 3D visualization in industrial digitalization are often understood separately. Engineers have difficulty seeing how a PLC signal becomes an application state, how equipment changes pass through each software layer, and whether the digital scene remains consistent with control behavior.
Solution
The system exposes the inputs and outputs of simulated PLC states through OPC UA, organizes equipment data through a REST service, and maps it to corresponding objects and behaviors in Unity. Unity interactions can change interface variables, which are transmitted through OPC UA to a PLC simulated with C# scripts for logic processing. The results return as Unity object behaviors, keeping the entire path clear and traceable.
Result / impact
The system organizes the complete path from industrial control logic to a visual digital scene as one coherent model, providing a compact reproduction of the framework used in current digital-twin systems.
Project evidence · Industrial Digital Twin Demo
Key features
Simulate equipment and control states to verify that the logic behaves correctly.
Transfer selected PLC data through a standard industrial connection.
Organize equipment information for use by software applications.
Reflect machine states and process changes in a Unity scene.
Observe the same state across control, service, and visualization layers.
Technical highlights
OPC UA exposes simulated PLC variables through a standardized industrial data model.
The REST layer separates industrial protocol handling from application-facing data access.
A shared state mapping connects PLC variables, service data, and Unity object behavior.
Unity components translate equipment states into visible motion, status, and process changes.
Layered responsibilities make signal paths easier to inspect and integration issues easier to locate.
The complete PLC → OPC UA → REST → Unity path provides a reusable reference for digital-twin integration.
Technologies
PLC
OPC UA
REST API
Unity
C#
Digital Twin
Simulation
Structure
Control path: PLC Logic Simulation → OPC UA Data Layer → REST Application Service → Unity State Mapping → Digital Twin Interface
Interaction path: Digital Twin Interface → Unity Interaction Variables → REST Application Service → OPC UA Data Layer → PLC Logic Simulation → State Feedback
BMW Group · Enterprise Project
Digital Twin Communication Plugin Integration
Integrate external-system communication into a digital-twin environment through a plugin, reducing additional software dependencies and improving model reuse across projects.
+
Project summary
A digital-twin integration project that packages external-system communication as a reusable plugin within the simulation environment, reducing additional dependencies while preserving the required system-integration capability.
Problem / overview
The existing digital-twin workflow relied on additional software for communication with external systems. This increased configuration and maintenance effort, added cost, and made digital models harder to reuse across projects.
Solution
The work covered analysis of the existing communication requirement, plugin-architecture design, development, and integration into the digital simulation environment. Encapsulating the required communication capability in the plugin reduced coupling between digital-twin models and the external implementation, supporting reuse across production-line projects and digital models.
Result / impact
The solution reduced reliance on additional software and its associated costs, while improving reuse across multiple production-line projects and related digital-twin models.
Key features
Package external-system communication as a plugin for the digital-twin environment.
Separate communication integration from project-specific digital-model logic.
Reuse the communication capability across production-line projects and digital models.
Deliver IT/OT integration with fewer additional software dependencies.
Technical highlights
Analyzed the existing communication requirement and designed the plugin architecture.
Developed and integrated the communication plugin within the digital simulation software.
Reduced coupling between digital-twin models and the external communication implementation.
Designed the integration for reuse across different models and production-line projects.
Technologies
Plugin Architecture
System Integration
Industrial Communication
IT/OT Integration
TUM · Personal Project
Multimodal 3D Navigation System
Use multimodal cues to guide users through complex spaces and toward their destination.
+
Project summary
A 3D navigation and interaction system that combines visual, auditory, and environmental interaction cues to improve immersion and perceptual quality in complex scenes.
Problem / overview
Traditional navigation in 3D scenes often relies on map-based prompts. This increases cognitive load and can disrupt the experience in scenes designed for immersion.
Solution
The navigation system combines scene interaction with changing streetlight intensity, footstep echoes, and firefly cues, helping users move progressively closer to the target.
Result / impact
The project improved user satisfaction by 30%, reduced cognitive load, and demonstrated that multimodal guidance can outperform conventional map guidance in specific scenarios.
Key features
Guide users with light, sound, and environmental changes instead of arrows alone.
Adapt the guidance to the user's current position and target location.
Use a human-factors evaluation framework to verify whether guidance reduces workload and improves spatial understanding.
Technical highlights
Navigation state combines current position, predicted position, and target point.
A dynamic triangular navigation region controls when environmental feedback is activated.
Streetlight intensity, footstep echo, and firefly guidance form complementary visual and auditory channels.
User evaluation and NASA-TLX analysis measure satisfaction and cognitive load.
Technologies
Unity
3D Interaction
Navigation Design
NASA-TLX
UX Evaluation
Multimodal Feedback
Structure
Algorithm: Actual User Position + Target Point + Environmental Trigger Point → Triangle Area Evaluation → Determine Interaction Trigger Direction
Interaction: User Movement Direction + Next Action → Select Environmental Node → Light / Sound / Environmental Feedback
ZF Group · Enterprise Project
Matrix-based Production Scheduling Model
Turn production conflicts into a computable scheduling model.
+
Project summary
A scheduling model based on weighted matrix logic that automatically generates production plans for customized product requirements.
Problem / overview
Production planning had to manage the conflict between standardized throughput and customized order requirements, but existing planning methods lacked a structured optimization model.
Solution
To resolve the conflict between the remaining workload of different orders and limited working hours, the model converts the production state into a remaining-workload matrix and a working-time distribution matrix. Weights then adjust the priority of customized demand before the model generates the corresponding production schedule.
Result / impact
The model provides a structured method for balancing efficiency and customization, validating an approach that turns scheduling decisions from experience-based judgment into computable logic.
Key features
Represent the remaining work for each production task.
Describe how available working time can be distributed.
Adjust priorities when customized orders require different treatment.
Generate a scheduling result from the combined production conditions.
Technical highlights
A remaining-workload matrix represents the current production state.
A workload-distribution matrix models available capacity allocation.
A configurable weight strategy converts customization constraints into scheduling priorities.
MATLAB provides the calculation environment for the initial scheduling model.
Technologies
MATLAB
Scheduling Model
Matrix Modeling
Weight Strategy
Production Planning
Esquel Group · Enterprise Project
Lean Process Optimization
Optimize lean production for a button manufacturing line.
+
Project summary
A process optimization project that uses value stream mapping to identify bottlenecks and non-value-added steps in button manufacturing, then applies established industrial-engineering methods to improve process layout and production efficiency.
Problem / overview
The button manufacturing process contained workflow waste, and actual production efficiency diverged from the overall capacity plan. Industrial-engineering methods were needed to identify bottlenecks and improve efficiency.
Solution
VSM was used to map the production process and locate problem points and bottlenecks. Video analysis and standard work-cell analysis then informed an optimization proposal for equipment layout and workflow.
Result / impact
The optimization proposal improved production efficiency, shifted the original bottleneck away from the critical production stage, and improved the capacity performance of the overall process.
Key features
Map the complete production process with VSM and identify bottlenecks.
Break down the actual production process through video analysis.
Use standard work-cell analysis to examine specific operations.
Optimize equipment layout and workflow based on the findings.
Technical highlights
Value stream mapping identifies bottlenecks and waste across the end-to-end production process.
Video analysis reveals additional efficiency issues in actual operations.
Standard work-cell analysis breaks down and compares specific production activities.
The findings are translated into equipment-layout and workflow optimization proposals.
Technologies
Lean
Value Stream Mapping
Video Analysis
Standard Work
Process Analysis
Industrial Engineering
04 / SKILLS
Range of skills
These capabilities come from practical work, research and personal exploration, often overlapping in different ways across projects.
ST
Systems Thinking
Start from the complete objective and understand the modules, information, interfaces and relationships within a system.
Areas involvedSystem Structure · Information Relationships · Interface Design · Problem Decomposition
