CNC tool-wear prognostics
A prognostics laboratory for estimating CNC cutting-tool wear from prepared vibration, acoustic, current, and machining-condition data.
Project definition
Problem statement
Tool wear changes cutting forces and sensor signals, but the same signals also vary with material, speed, feed, depth of cut, tool geometry, and sensor mounting.
The engineering problem is to estimate wear or remaining useful life without leaking information between portions of the same machining run and to state uncertainty when operating conditions differ from training data.
Project objectives
- Prepare synchronised machining conditions, sensor windows, tool identity, cut count, and wear measurements.
- Extract time, frequency, and time-frequency features from selected signals.
- Compare threshold, regression, classification, and sequence-based baselines where appropriate.
- Separate complete tool runs between training, validation, and testing.
- Measure wear error, maintenance-threshold detection, remaining-life error, and confidence reliability.
System design
System modules
Run data manager
Stores tool, workpiece, cutting conditions, sensor channels, sampling metadata, wear labels, and run splits.
Signal processor
Synchronises, filters, segments, and checks vibration, acoustic, current, and related signals.
Feature engine
Calculates statistical, spectral, band-energy, envelope, and selected time-frequency features.
Prognostic model
Estimates wear state or remaining life with run-separated training and saved configurations.
Maintenance dashboard
Shows tool history, predictions, uncertainty, thresholds, errors, and operating-condition differences.
Methodology
System workflow
- 01Load machining runs
The student selects prepared runs and verifies tool, material, cutting, sensor, and wear metadata.
- 02Prepare windows
Signals are quality-checked, synchronised, segmented, and assigned by complete run to fixed splits.
- 03Train models
Baselines and selected prognostic models use the same training runs and features.
- 04Predict wear
Test runs are processed in time order without access to later wear values.
- 05Review maintenance point
Prediction error, confidence, and threshold timing are compared with measured wear.
Demonstration scenario
Several prepared milling runs train two wear models, while an entire tool run remains unseen. The dashboard replays the unseen run, compares predicted and measured flank wear, marks the maintenance threshold, and shows increased uncertainty when the cutting conditions move outside the training range.
Engineering
Technical architecture
- Web application
- Nuxt and Vue for run selection, signal plots, feature views, wear curves, thresholds, and comparisons.
- Prognostics API
- FastAPI for datasets, signal jobs, features, models, predictions, evaluation, and exports.
- Data layer
- PostgreSQL for runs, conditions, window metadata, features, wear labels, models, and predictions.
- Signal pipeline
- Python, NumPy, and SciPy for filtering, segmentation, spectral analysis, and feature extraction.
- Model pipeline
- scikit-learn and selected sequence models with group-based splits, uncertainty estimates, and reproducible seeds.
Testing
Evaluation
Evaluation measures
- Mean absolute and root mean squared wear error
- Maintenance-threshold precision, recall, and timing error
- Remaining-useful-life error where run-to-failure labels are available
- Performance on tools and cutting conditions excluded from training
- Confidence calibration and out-of-distribution warning behaviour
- Feature, training, and inference time across signal volumes
System boundaries
- Predictions apply only to the documented tool, material, machine, sensors, and operating ranges.
- The system does not control a CNC machine or stop a machining process.
- A maintenance prediction does not replace inspection or manufacturer guidance.
- Public, synthetic, or explicitly permitted machining data is used.
Included
- 01Machining run and sensor-analysis interface
- 02Signal, feature, wear, and remaining-life pipelines
- 03Wear trend, confidence, error, and maintenance dashboards
- 04Prepared machining datasets, experiments, and test results
- 05Complete source code in a private GitHub repository
- 06Complete project documentation with synopsis, abstract, methodology, machining and signal-flow diagrams, experiment results, screenshots, and conclusion
- 07Setup and usage guide
Project record
No buyer information is collected on this page.
- Permanent project ID
- GP-ME-1JLTYPI
- Catalogued
- 21 Aug 2026
- Completed
- Pending
- Verified
- Pending
- Demonstration
- Added when ready
Handover
After purchase
- 01Payment is confirmed
The project is marked unavailable and cannot be purchased again.
- 02Repository access is granted
The buyer's submitted GitHub account receives access to the private repository.
- 03The purchase record is delivered
The certification sheet is prepared from the reviewed buyer details and sent privately by email.