Cyclic Manufacturing Monitoring Using ML Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing manufacturing systems lack effective data collection and predictive methods to detect wear and tear in manufacturing devices, leading to unplanned downtimes and high operational costs due to unforeseen failures.
Innovation Solution
Implement a computer-implemented method using a trained machine learning model to measure and compare actual operation values with reference values, employing a distance function to initiate alerts when deviations exceed predetermined thresholds, thereby optimizing maintenance scheduling and reducing downtime.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If regular maintenance intervals are used, then maintenance scheduling is simple, but unplanned downtime increases due to undetected wear and tear
Solution Approach 1:
The system performs preliminary detection of anomalies by continuously monitoring physical properties and comparing them against learned normal patterns. This early detection allows maintenance to be scheduled before failures occur, transitioning from reactive to proactive maintenance and reducing unplanned downtime.
Solution Approach 2:
The patent replaces traditional mechanical monitoring approaches with a data-driven machine learning system. Instead of relying on fixed maintenance schedules or simple threshold-based alarms, the system uses neural networks to learn complex patterns from sensor data, enabling more accurate and adaptive anomaly detection.
2Productivity
If fixed maintenance intervals are applied, then maintenance planning is straightforward, but operational costs increase due to unnecessary maintenance and unplanned failures
Solution Approach 1:
The system establishes a continuous feedback loop where sensor data from the manufacturing device is constantly collected, analyzed by the machine learning model, and used to update the maintenance schedule. This feedback mechanism ensures maintenance is performed based on actual device condition rather than arbitrary time intervals, optimizing both cost and reliability.
Solution Approach 2:
The system changes the maintenance parameter from fixed time intervals to condition-based thresholds. By monitoring physical properties and comparing them against dynamically determined reference values, the system adapts maintenance timing to actual device degradation patterns, reducing unnecessary maintenance while preventing failures.
3Measurement precision
If no monitoring system is implemented, then system complexity remains low, but detection of wear and tear anomalies is impossible
Solution Approach 1:
The monitoring system is segmented into distinct functional modules: sensor data acquisition, preprocessing, machine learning inference, and alert generation. This modular architecture manages complexity by separating concerns, allowing each component to be optimized independently while maintaining high measurement precision through specialized processing at each stage.
Data Source
AI summary
A computer-implemented method of monitoring a cyclically operating manufacturing device includes measuring actual values of a physical property relating to operation of the manufacturing device during multiple cycles of the operation of the manufacturing device. Reference values are determined for the physical property for each of the multiple cycles based on a trained machine learning model. A distribution of the actual values is compared with a distribution of the reference values based on a distance function, and an alert is initiated when the distance function exceeds a predetermined threshold.