Embedded signal fault-detection laboratory
An embedded signal-processing laboratory for detecting sensor faults under memory, latency, and compute constraints using prepared time-series data.
Project definition
Problem statement
Sensor faults such as bias, drift, spikes, noise increase, clipping, and disconnection can corrupt an embedded control or monitoring system.
The engineering problem is to detect selected faults from streaming signals while meeting realistic limits on memory, processing time, sampling rate, and false alarms.
Project objectives
- Prepare versioned normal and faulty sensor time series with documented sampling conditions.
- Implement rule-based, statistical, and lightweight learned fault detectors.
- Replay signals as streams and detect faults within a measured delay.
- Profile model size, working memory, operation count, and execution latency.
- Compare detection accuracy with embedded resource use and false alarms.
System design
System modules
Signal manager
Imports signals, validates sampling metadata, creates fixed splits, and stores fault labels.
Fault injector
Adds reproducible bias, drift, spike, noise, clipping, dropout, and stuck-value faults.
Feature pipeline
Calculates time and frequency features using bounded streaming windows.
Detector workbench
Runs thresholds, statistical detectors, and selected compact models through one interface.
Embedded profiler
Measures detection delay, model size, estimated memory, operations, and target or emulator latency.
Methodology
System workflow
- 01Select signal
The student chooses a prepared sensor stream, sample rate, window, and detector.
- 02Create scenario
A documented fault type, magnitude, start time, and duration are applied.
- 03Train or configure
Detector parameters use only the training portion of the signal set.
- 04Replay stream
Samples are processed in order under the selected resource and timing settings.
- 05Compare tradeoffs
Accuracy, false alarms, delay, memory, and latency are compared across detectors.
Demonstration scenario
A vibration signal is replayed normally and then with gradual sensor drift and short spikes. Three detectors process identical windows. The dashboard compares the first detection time, false alarms, memory, and latency, then shows the selected lightweight detector running through the embedded interface.
Engineering
Technical architecture
- Web application
- Nuxt and Vue for signal plots, fault controls, detector setup, timelines, and tradeoff comparison.
- Experiment API
- FastAPI for datasets, fault generation, detector jobs, profiles, and exports.
- Signal pipeline
- Python for offline preparation and evaluation, with fixed-window processing matching the embedded implementation.
- Embedded implementation
- C or TensorFlow Lite components for selected features and compact detector inference.
- Reproducibility
- Versioned datasets, deterministic fault seeds, saved configurations, prepared targets, and regression tests.
Testing
Evaluation
Evaluation measures
- Precision, recall, F1 score, and false alarms per operating hour
- Detection delay by fault type and magnitude
- Robustness to normal operating changes and sensor noise
- Agreement between offline and embedded inference outputs
- Model size, working memory, operation count, and measured latency
- Maximum sustainable sample rate under prepared target conditions
System boundaries
- The project detects only prepared sensor and fault types under documented operating conditions.
- Detected anomalies do not identify the physical cause without engineering inspection.
- The prototype does not control safety-critical equipment.
- Resource measurements apply to the included target or emulator configuration.
Included
- 01Signal dataset and fault-scenario interface
- 02Feature, detector, replay, and embedded-profiling modules
- 03Latency, memory, accuracy, and error-analysis dashboard
- 04Prepared sensor signals, faults, models, and test results
- 05Complete source code in a private GitHub repository
- 06Complete project documentation with synopsis, abstract, methodology, 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-EC-1PRZ799
- 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.