Health Index

An effective Health Index synthesizes heterogeneous data streams into a single coherent value. It does not rely on a single sensor type.

• Vibration Signatures: Captures mechanical looseness and imbalance.
• Thermography: Detects friction-induced heat anomalies.
• Oil Debris Analysis: Quantifies metallic particle concentration for wear.
• Motor Current Signature Analysis (MCSA): Identifies electrical faults. By fusing these modalities, the index eliminates blind spots inherent in single-sensor monitoring.

The index must map complex physical phenomena to a universal, intuitive scale, typically 0.0 to 1.0 or 0% to 100%.

• 1.0 (100%): Represents a pristine, factory-new state.
• 0.0 (0%): Defines the functional failure threshold. This normalization abstracts away raw engineering units (e.g., mm/s, °C) into a business-logic metric that triggers maintenance workflows uniformly across diverse asset fleets.

A reliable Health Index exhibits a monotonically decreasing trend over the asset's lifecycle, barring maintenance interventions. It should not fluctuate wildly due to transient operational noise.

• Noise Filtering: Employs exponential smoothing or Kalman filters to suppress process noise.
• Trend Consistency: A sudden increase in health (without repair) indicates a flawed model. This property ensures that the Remaining Useful Life (RUL) curve derived from the index is predictable and trustworthy.

The index must be sensitive enough to detect incipient faults long before they cascade into catastrophic failures. It should deviate from the baseline while the asset is still in the P-F interval (Potential Failure to Functional Failure).

• High Sensitivity: Detects subtle changes in kurtosis or skewness of vibration signals.
• Lead Time: Provides sufficient warning for planned parts procurement and scheduling. This shifts the maintenance strategy from reactive to Condition-Based Maintenance (CBM).

While the Health Index is a composite, its underlying feature set must correlate with specific Failure Mode Classifications. A dropping index should be explainable.

• Bearing Wear: Driven by increased high-frequency vibration energy.
• Misalignment: Driven by elevated 1x and 2x running speed harmonics.
• Cavitation: Driven by specific pressure pulsation patterns. Using SHapley Additive exPlanations (SHAP) values, engineers can decompose the index drop to identify the dominant physical failure driver.

A static threshold is insufficient. An effective Health Index accounts for the asset's operational regime to avoid false positives.

• Load Normalization: Distinguishes between high-vibration due to heavy load versus actual degradation.
• Speed Compensation: Adjusts fault frequencies relative to variable speed drives.
• Environmental Factors: Compensates for ambient temperature fluctuations. Without context awareness, the index generates nuisance alerts that erode operator trust.

The predicted duration until an asset reaches its functional failure threshold. While the Health Index provides a normalized snapshot of current degradation, RUL translates that index into a concrete time-based forecast. A Health Index of 0.2 might correspond to an RUL of 30 days, enabling precise scheduling. RUL estimation relies heavily on time-series forecasting and degradation modeling to map the current state to a future failure point.

The algorithmic process of combining data from disparate sources to create a unified, more accurate state estimate. A Health Index is fundamentally a product of sensor fusion, integrating inputs like vibration, temperature, and acoustics. By weighting and correlating these streams, the system reduces uncertainty and noise. For example, a spike in temperature alone might be a false positive, but when fused with a specific Fast Fourier Transform (FFT) signature, it confirms a bearing fault.

A maintenance strategy triggered by the actual asset condition rather than a fixed schedule. The Health Index serves as the primary trigger for CBM actions. When the index crosses a predefined threshold—say, dropping below 0.6—a work order is automatically generated. This contrasts with reactive maintenance (run-to-failure) and preventive maintenance (time-based), optimizing resource allocation by servicing only equipment that shows objective evidence of degradation.

The overarching engineering discipline that encompasses the entire lifecycle of asset health. The Health Index is the core quantitative output of a PHM system's diagnostic and prognostic modules. PHM frameworks standardize how raw sensor data is transformed into actionable intelligence, covering:

• Diagnostics: Detecting and isolating a fault (using anomaly detection).
• Prognostics: Predicting future degradation (calculating the Health Index and RUL).
• Prescriptive Maintenance: Recommending specific repair actions based on the prognosis.

The domain-expertise-driven process of extracting meaningful statistical attributes from raw sensor streams to feed into a Health Index model. Raw vibration data is useless without transformation. Key features include root mean square (RMS) for overall energy, kurtosis for impulsiveness, and specific frequency band amplitudes from an FFT. The quality of a Health Index is directly dependent on the selection of features that correlate most strongly with physical wear mechanisms like spalling or pitting.

The mathematical representation of how a system's health deteriorates over time. This model forms the backbone of the Health Index trajectory. Common models include:

• Exponential: For accelerated wear like chemical corrosion.
• Polynomial: For gradual, linear wear like abrasive friction.
• Piecewise: For assets with distinct run-in, steady-state, and wear-out phases. The model is fit to historical run-to-failure data to predict how the Health Index will decay from its current value to zero.