What makes a sensor truly smart? It’s not just the ability to collect data—but the power to interpret, process, and act on that data autonomously. At Delphisonic, we’ve engineered our DS-Track sensor architecture to do precisely that: integrate robust hardware, embedded intelligence, and AI algorithms into a single, compact system optimized for predictive maintenance.
This blog takes you inside the brain of our sensor—unpacking the physical layers, the data pipelines, and the signal processing logic that make Delphisonic a leader in edge-based fault detection.
1. Hardware Foundation: Designed for Extremes
At the heart of our system is a rugged yet highly precise sensor platform engineered for the most demanding industrial conditions.
Key hardware components:
- Analog MEMS Accelerometers (e.g., ADXL1002 / ADXL1004):
- ±500gE to ±1000gE range
- Extremely low noise density (<25 µg/√Hz)
- High shock survivability (10,000g)
- Temperature Sensor (±0.5°C accuracy)
- Microcontroller (ARM Cortex-M Series):
- High-speed ADC (Analog-to-Digital Conversion)
- DSP engine support
- On-board RAM for buffering
- Industrial Enclosure:
- IP69-rated for water, dust, and vibration protection
- -40°C to +120°C operating range
- Stainless steel or hard-anodized aluminum casing
- Connectivity Modules:
- Wi-Fi / BLE / LoRa / NB-IoT
- Optional Single Pair Ethernet (SPE) for data + power
This foundation allows DS-Track to operate on train bogies, traction motors, gearboxes, or any high-stress machinery with unwavering reliability.
2. Embedded Signal Processing: Intelligence at the Edge
Raw data alone doesn’t provide insight. That’s why DS-Track performs real-time embedded signal processing directly inside the sensor.
Key algorithms executed on-device:
- Band-Pass Filtering: Removes irrelevant low- and high-frequency noise to isolate fault frequencies.
- Hilbert Transform + Envelope Extraction: Converts raw acceleration into amplitude-modulated waveforms to highlight bearing faults and resonance signatures.
- Fast Fourier Transform (FFT): Transforms time-domain data into frequency domain for identifying characteristic fault frequencies (e.g., BPFO, BPFI, BSF).
- Root Mean Square (RMS) and Crest Factor Calculation: Provides trendable health indicators for monitoring wear progression.
These algorithms are optimized for low power consumption and real-time execution—delivering actionable insights in less than 5 milliseconds.
3. AI-Based Fault Detection: Learning from Machines
Beyond signal processing, Delphisonic’s proprietary AI engine leverages years of labeled vibration data to identify, classify, and predict failure patterns.
Our AI layer includes:
- Pattern Recognition Models: Trained on over 570 known fault signatures, from bearing pitting to gear misalignment.
- Threshold Adaptation Algorithms: Continuously adjust normal operating ranges based on machine learning and environment context.
- Anomaly Scoring: Assigns confidence levels to deviations using statistical learning and frequency pattern deviation.
- Embedded Decision Tree or SVM Models (on sensor): Allows each device to autonomously determine:
- Whether a fault exists
- Its severity level (Green, Yellow, Red)
- If immediate alerting is required
This edge-AI model can operate without needing constant cloud support, a must-have in remote or mobile applications.
4. Communication & Integration: Flexible by Design
Once processed, only relevant data and fault events are transmitted, significantly reducing bandwidth usage.
- Protocol Support:
- MQTT, HTTP(S), Modbus-TCP, REST API
- Data Transfer Options:
- Local gateway → cloud
- Direct NB-IoT to cloud
- Single Pair Ethernet to on-board control unit
- Dashboard Integration: Compatible with custom UI or existing platforms like SCADA, CMMS, or fleet analytics dashboards.
Thanks to its modular design, DS-Track can be easily integrated into trains, metro systems, industrial pumps, or wind turbine monitoring systems.
5. Example Application: Train Axle Box Monitoring
In one of our recent deployments:
- 8 DS-Track sensors were mounted on each locomotive’s axle boxes
- Signal processing + fault detection ran entirely on-device
- Edge results were transmitted once per event via NB-IoT
Result: A bearing defect that would have led to a hotbox event was detected 13 days in advance, enabling scheduled maintenance and avoiding service disruption.
6. System Benefits at a Glance
| Feature | Value |
| Sampling Rate | Up to 50kHz |
| Sensing Range | ±1gE to ±1000gE |
| Edge AI Fault Classes | 570+ patterns |
| Alert Delay | <5 ms |
| Power Supply | DC, Battery, or PoE |
| Data Output | Raw + Processed + AI Result |
Conclusion
Delphisonic’s sensor architecture is more than just data collection—it’s autonomous, intelligent condition monitoring engineered from the ground up for reliability, accuracy, and adaptability.
In a world where downtime costs millions and safety is paramount, smart sensors like DS-Track are redefining how industries look at health monitoring.
Want to explore how this architecture fits your assets? Let’s talk.
