Vertical Machining Center (CNC)

• ML model for thermal error prediction and compensation for Spindle position
• Real-time temperature sensor and Fanuc controller parameters data is collected through edge device
• ML model process the data in real-time on the edge device and predicts error
• Then sends the error correction to Fanuc controller. Real time temperature and AI predicted correction values are also displayed on HMI screen.

• 80% reduction in thermal positioning error (Z-axis: 69µm → 13µm; Y-axis: 27µm → 3µm)
• ML model provides correction for every ±0.15°C temperature change
• Minimum error prediction accuracy of ±0.1µm

Vertical Machining Center (CNC) — Stainless Steel, Steel, and Cast Iron applications

• ML model for tool/insert wear condition prediction, edge deployment pipe-line, MQTT protocol for data/write with Fanuc controller.
• Real-time controller (Fanuc) parameters data is collected through edge device; ML model process the data in real-time on the edge device and classifies tool/insert condition.
• Tool/insert condition classified in 3 categories: Non-wear, wear and catastrophic/severe.
• Real time AI prediction of tool/insert condition is displayed on HMI screen for end-user to take an action.

• 82% accurate prediction of tool wear condition SS, Steel and Cast Iron

• GenAI based Intelligent Service system deployed on AWS
• GenAI based RAG Architecture
• Re-ranking module, Backend Module, Database, Frontend Module
• APIs for accessing live SR ticket data, APIs for accessing closed ticket SR, DEMS data.

• RAG (Retrieve and Generate) approach uses the historical SR, DEMS, LMW Manuals, WhatsApp data to find the most relevant solutions for a live ticket issue.
• When a service engineer asks a question, the AI solution retrieves similar cases from past tickets and provides multiple solution suggestions.

• Chatbot application on LMW AWS Cloud
• Role based access for users: Admin/HO and LMW’s Service Engineer
• Application has Admin page for performance monitoring and Service Engineer page has chatbot feature.

• Web app dashboard with integrated chatbot (AWS-based deployment)
• Predict yarn quality based on input quality and machine settings
• Predict machine settings and input quality based on required output quality

• Collects raw material properties, machine settings, and process parameters
• Predicts yarn quality metrics (CSP, RKM, U%, IPI) using trained ML models
• Provides recommendations for optimal machine settings and raw material selection
• Accessible via chatbot for both technical and non-technical users

• Forward Prediction: 90.48% accuracy
• Reverse Prediction: 90% accuracy
• Out of 13 critical yarn parameters, 11 parameters observed significant correlation

Finisher Draw Frame (Textile Spinning)

• 82% accuracy in detecting true fault alarms
• ~93% reduction in nuisance stoppages vs. legacy “always-stop” logic
• 128ms average inference latency — well within the 400ms operational budget

• Built an ML model trained on historical alarm logs from PLC and SpinConnect, labeling alarms as True (critical) or False (nuisance) using rule-based logic and operator sequences
• Deployed on an edge device with OPC-UA integration — classifying each alarm in real time and sending automated severity commands (STOP, WARN, or IGNORE) directly to the PLC

Vertical Machining Center (CNC) — Stainless Steel, Steel, and Cast Iron applications

• ML model for tool/insert wear condition prediction, edge deployment pipe-line, MQTT protocol for data/write with Fanuc controller.
• Real-time controller (Fanuc) parameters data is collected through edge device; ML model process the data in real-time on the edge device and classifies tool/insert condition.
• Tool/insert condition classified in 3 categories: Non-wear, wear and catastrophic/severe.
• Real time AI prediction of tool/insert condition is displayed on HMI screen for end-user to take an action.

• 82% accurate prediction of tool wear condition SS, Steel and Cast Iron

Finisher Draw Frame — 1 Servo & 2 Servo configurations

• Root cause identified: 65–100ms roller startup lag and 36:1 draft ratio surge
• Corrective solution validated: EMA-based voltage correction + multi-segment ramping

• Built a diagnostic analysis pipeline capturing high-resolution RPM and draft ratio signals via encoders and LVDTs — identifying a 65–100ms timing lag between front and middle rollers causing transient tension imbalance and sliver thinning
• Diagnosed draft ratio surging to 36:1 at startup, causing 6–8m mass dips; explored multi-segment acceleration ramps, PID retuning, and EMA-based voltage correction to eliminate overshoot and sensor drift
• Delivered as upgraded PLC logic with multi-segment ramping and PID tuning for smooth roller synchronization across Acc/Dec cycles