Machine Acoustic Analysis Threshold Learning for Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine monitoring systems face challenges in detecting anomalies with limited training data, leading to suboptimal determination of operating states due to insufficient statistics for abnormal operation, resulting in reduced quality of anomaly detection and increased false detection rates.
Innovation Solution
A method that involves acquiring and processing signals from sensors to calculate deviation signals by subtracting characteristic signals of normal operation from those of abnormal operation, thereby increasing the number of signals used to determine thresholds, which are then recorded in a reference database for improved anomaly detection during the test phase.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If statistical analysis is performed with limited training data, then the learning process can be completed, but the determined thresholds are not optimized leading to reduced detection accuracy
Solution Approach 1:
The patent creates artificial copies of training data by generating deviation signals through subtraction operations. For each characteristic signal of normal operation, multiple deviation signals are created by subtracting it from other characteristic signals (both normal and abnormal). This copying approach multiplies the effective training samples without requiring additional physical training data, thereby improving threshold optimization and detection accuracy despite limited original training data availability.
2Measurement precision
If more training data is collected to improve statistical analysis, then detection accuracy improves, but the complexity and time of the learning phase increases
Solution Approach 1:
The patent performs preliminary actions by pre-calculating and storing characteristic signals of normal and abnormal operation during the learning phase. These characteristic signals are then reused multiple times to generate deviation signals through subtraction operations. This preliminary preparation eliminates the need to process raw training data repeatedly, significantly reducing the time required for threshold determination while maintaining high detection accuracy.
3Reliability
If traditional threshold determination methods are used with limited data, then the process remains simple, but false detection rates increase
Solution Approach 1:
The patent introduces deviation signals as an intermediary element between raw characteristic signals and final threshold determination. These deviation signals, obtained through subtraction operations, serve as mediators that highlight the differences between normal and abnormal operations. This intermediary approach improves detection reliability by focusing analysis on meaningful variations while maintaining manageable processing complexity through systematic signal generation and filtering.
Data Source
Figure 1
Figure 2
Figure 3a~3b
AI summary
The invention pertains to a method of analysis of the state of operation of a machine (M) comprising a learning step supplementing a reference database (3) with one or more thresholds for one or more indicators calculated on the basis of signals delivered by a sensor (2) associated with the machine, the learning step comprising the following operations implemented by a computer processing unit (10): - an acquisition of signals characteristic of normal operation and of abnormal operation of the machine; - of each of the signals characteristic of normal operation, formation of at least one so-called deviation signal by implementing a mathematical operation having as attributes the signal characteristic of normal operation and one of the signals characteristic of normal or abnormal operation other than said signal characteristic of the normal operation; - for each of the deviation signals, calculation of an indicator; - determination of an indicator threshold representative of a limit between normal operation and abnormal operation of the machine.