Predictive Maintenance Baseline Model for Fault Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional predictive maintenance methods struggle to accurately detect faults in production equipment due to the influence of multiple parameters, leading to improper scheduling and increased maintenance costs or reduced production capacity.
Innovation Solution
A baseline predictive maintenance method using virtual metrology that computes a healthy baseline value and device health index to infer the status of a target device and forecast remaining useful life, incorporating a TD baseline model, device health index, baseline error index, and individual similarity indices to achieve fault detection and classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Difficulty of detecting and measuring
If conventional SPC approach is used to monitor key parameters, then simple monitoring is achieved, but fault detection accuracy deteriorates due to inability to account for inter-parameter influences
Solution Approach 1:
The patent segments the monitoring system into multiple individual models, each dedicated to predicting a specific key parameter. Instead of using a single complex multivariate model, the system creates separate predictive models for each parameter (e.g., throttle-valve angle model, ammonia concentration model), allowing independent analysis of each parameter's deviation from expected behavior based on related parameters.
Solution Approach 2:
The patent introduces related parameters as intermediary variables that mediate the relationship between key parameters and their deviations. By modeling how related parameters influence key parameters, the system can distinguish between deviations caused by the key parameter itself versus those caused by changes in related parameters, thereby improving fault detection accuracy without requiring a single complex model.
2Ease of operation
If scheduled preventive maintenance is performed at predetermined intervals, then maintenance planning is simplified, but production capacity deteriorates due to improper scheduling
Solution Approach 1:
The patent implements a feedback mechanism where the individual predictive models continuously monitor key parameters and generate alerts when deviations exceed predefined thresholds. This feedback loop enables dynamic adjustment of maintenance scheduling based on actual equipment condition rather than fixed intervals, allowing maintenance to be performed only when necessary, thus preserving production capacity while maintaining operational simplicity.
Solution Approach 2:
The patent transitions from static predetermined maintenance intervals to dynamic condition-based maintenance scheduling. By using real-time monitoring of key parameters and their deviations predicted by individual models, the system adapts maintenance timing to actual equipment state, performing maintenance earlier when degradation is detected and later when equipment remains healthy, thereby optimizing production capacity.
3Measurement precision
If multiple parameters are monitored simultaneously, then comprehensive monitoring is achieved, but fault detection accuracy deteriorates due to inability to distinguish causal relationships
Solution Approach 1:
The patent segments the monitoring of multiple parameters into separate individual predictive models, each focusing on one key parameter. This segmentation allows the system to handle multiple parameters without the complexity of multivariate analysis, as each model independently predicts its designated parameter based on related parameters, maintaining measurement precision while avoiding the complexity of simultaneous multi-parameter analysis.
Solution Approach 2:
The patent applies local quality by tailoring each individual predictive model to the specific characteristics and relationships of its designated key parameter. Each model is optimized for its specific parameter with customized related parameters and prediction algorithms, allowing precise fault detection for each parameter while keeping individual model complexity manageable, rather than using a single complex model for all parameters.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
A baseline predictive maintenance method for a target device (TD) and a computer program product thereof are provided. Fresh samples which are generated when the target device produces workpieces just after maintenance are collected, and a new workpiece sample which is generated when the target device produces a new workpiece is collected. A plurality of modeling samples are used to build a TD baseline model in accordance with a conjecturing algorithm, wherein the modeling samples include the new workpiece sample and the fresh samples. A TD healthy baseline value for the new workpiece is computed by the TD baseline model, and a device health index (DHI), a baseline error index (BEI) and baseline individual similarity indices (ISIB) are computed, thereby achieving the goals of fault detection and classification (FDC) and predictive maintenance (PdM).