Digital Bearing Fingerprint: New AI Method That Sees What No Sensor Can
Modern industry is dominated by an approach based on threshold control and analysis of individual parameters (RPM, vibration, temperature, pressure). These methods are important but suffer from a critical limitation: they are reactive.
An alarm is triggered when the fault has already become apparent and the parameter has crossed a hard limit. They answer the question “What broke?” — but almost always too late.
Modern systems require a proactive approach — the ability to detect an emerging problem in a complex web of interdependencies rather than in a single isolated indicator.
2. The Idea of a Fingerprint Analysis
To address this, we propose a shift from data analysis to state analysis. The new approach is based on Fingerprint Analysis (FP-analysis) — a technology that transforms disparate data streams into a unified, information-rich state signature of the equipment.
2.1 Step 1: Contextual Encoding — Creating a Parameter “Fingerprint”
Each individual parameter (temperature, pressure and etc) is not just captured as a number but is transformed using AI-based methods into its own complex, multidimensional fingerprint representing the system’s state.
This unique state preserves not only the parameter’s value but also its context:
- Dynamics: How fast is it changing?
- Trends: Is it rising, falling, or fluctuating?
- Correlations: How does its change correlate with vibration, pressure, and other parameters?
The fingerprint allows the integration of different types of physical parameters into a single state signature, where even a small but unusual change — invisible to a threshold system — leaves a unique trace.
2.2 Step 2: Aggregation — Forming a Subsystem “Signature”
Individual fingerprints (e.g., viscosity, pressure, contamination level) are aggregated into a single, comprehensive vector — a functional subsystem signature (e.g., “lubrication system signature”). This signature represents a comprehensive “digital DNA” of a component’s state at a given moment.
2.3 Step 3: Approach Validation on a Digital Twin
To build and validate such a sensitive approach, reliable data is required — which is difficult to obtain in the real world. Therefore, the system prototype was developed and validated using a complex software simulator — a digital twin of a bearing assembly.
This emulator reproduced dozens of interconnected physical processes, from vibration and friction to changes in temperature, lubricant viscosity, and wear level.
The model accounted for nonlinear dependencies — for example: temperature rise accelerates lubricant degradation, which increases friction and causes further heating.
This complex and dynamic environment provided ideal conditions to prove that FP-analysis can detect the earliest and weakest stages of degradation that traditional methods miss.
3. How AI Technology Becomes an Understandable Solution
The ability to analyze complex signatures, verified on the simulator, enables the system to act as a smart assistant. Deviation of the current signature from the reference is not just an “ALERT” but an informative trigger for deep diagnostics. Accurate issue description is achieved through integration with LLM models (e.g., ChatGPT or local LLM engine), which can formulate the problem in simple and understandable language for the maintenance staff.
Get Kirill | AI Solutions for High-Tech’s stories in your inbox
Join Medium for free to get updates from this writer.
Example: The system detects that the current “lubrication system signature” has shifted towards a pattern typical of increased friction and thermal degradation. No individual parameter has yet exceeded the norm, but the overall picture has changed. Based on this analysis, the system generates not an error code but a clear diagnosis:
🔍 DETECTED: The lubrication system is operating under increased load.
🔧 CAUSE: Likely lubricant thickening due to contamination build-up (inferred from signature correlations).
⚠️ RISKS: If no action is taken within a week, accelerated bearing wear is likely.
✅ ACTION: Check and clean lubrication filters.
4. Key Advantages of Fingerprint Analysis
This approach delivers quality benefits unattainable with traditional systems:
- Preventive Detection:
The system detects deviations weeks or months before parameters exceed limits. It sees the emergence of the issue, not just the issue itself. - Functional Localization:
The analysis identifies exactly which subsystem (cooling, lubrication, power supply) is the source of the problem — not just reports a specific sensor’s deviation. - Universality and Scalability:
The approach is applicable to any equipment type where multichannel data is available — from pumps to complex production lines. - Interpretability:
Despite the complexity “under the hood,” results are easily visualized and translated into clear language, offering specific recommendations, forecasts, and risk assessments.
5. Conclusion
Fingerprint Analysis is a fundamental rethinking of the technical condition monitoring paradigm. It shifts the focus from reactive fault tracking to proactive “health” management of equipment. Instead of waiting for a red light to go on, we gain the ability to see how and why it may light up in the future — and prevent it at minimal cost.
Additional Materials:
Below are several operational Grafana dashboards with real-time calculations and indicators for various subsystems based on simulator data.
Sensor data from various subsystems in our example is grouped, allowing the monitoring of a single system module fingerprint rather than many parameters, with the ability to drill down when a deviation occurs.
For example, sensor groups include:
all_temperatures– temperature sensors setall_vibrations– vibration sensors setall_viscosity– lubricant viscosity estimatesall_contamination– bearing contamination levelsall_oil_temperature– oil temperature sensors setall_pressure– oil pressure system
Telemetry parameters can be analyzed in any combination as needed for business purposes.
Additional simulation information:
- The simulation includes 5 bearing blocks, each with 15 sensors
- Each sensor transmits data twice per second
- Analysis interval is 10 seconds
- System behavior prediction is configured for +10 intervals (100 seconds) into the future
Thank you for your attention! The world of technology does not stand still, and we are happy to share our developments with you. Follow the news!
