Manufacturing Waveform Anomaly Detection With Unsupervised Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing manufacturing anomaly detection methods face challenges in effectively capturing every anomaly due to the vast number of tests performed on complex items, making it difficult to design suitable pass and/or fail conditions.
Innovation Solution
The system utilizes unsupervised machine learning to detect manufacturing anomalies by training a model on a dataset of component waveforms, refining the output through a feedback loop, and applying clustering algorithms to identify anomalies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a vast number of tests are performed on complex items to ensure quality, then measurement precision and reliability improve, but device complexity and difficulty of detecting and measuring increase
Solution Approach 1:
The patent introduces unsupervised machine learning models as intermediaries between the test data and anomaly detection process. These models automatically learn patterns from component waveforms and test responses, serving as a mediator that simplifies the complex task of anomaly detection across thousands of tests without requiring manual configuration of pass/fail conditions for each test parameter.
Solution Approach 2:
The patent replaces traditional mechanical rule-based anomaly detection systems with data-driven machine learning models. Instead of manually setting thresholds and rules for each of the 7000+ tests, the system uses unsupervised learning algorithms to automatically identify anomalies, substituting complex mechanical configuration with automated computational analysis.
2Measurement precision
If manual pass and fail conditions are designed for each test, then measurement precision improves, but ease of manufacture and productivity deteriorate
Solution Approach 1:
The unsupervised machine learning models perform self-service by automatically learning anomaly patterns from training data without requiring manual intervention for each test condition. The models autonomously configure detection parameters and make real-time anomaly decisions, eliminating the need for human experts to manually design pass/fail conditions for thousands of tests while maintaining high detection accuracy.
Solution Approach 2:
The system performs preliminary action by training the unsupervised learning models on historical test data before actual production testing. This pre-training phase allows the models to learn normal and anomalous patterns in advance, enabling rapid real-time anomaly detection during manufacturing without requiring manual rule configuration for each new product or test scenario.
3Reliability
If more tests are performed to capture every anomaly, then reliability improves, but loss of time and productivity increase
Solution Approach 1:
The patent extracts the essential anomaly detection function from the vast number of individual tests by using unsupervised learning models to identify patterns across multiple test parameters simultaneously. Instead of evaluating each of the 7000+ tests separately with manual rules, the system extracts meaningful anomaly signals by learning correlations across the entire test dataset, reducing the effective detection burden while maintaining comprehensive quality assurance.
Data Source
AI summary
Systems and methods are described for training a model for detecting manufacturing anomalies. A test response parameter is identified at a computing device, and a first plurality of component waveforms associated with the test response parameter are received at the computing device. Each waveform of the plurality of waveforms comprises a plurality of datapoints. A model is generated at the computing device, and the model is trained at the computing device and on the first plurality of component waveforms, thereby generating one or more parameters associated with the model. A second plurality of component waveforms associated with the test response parameter is received, and the trained model is accessed. It is indicated using the trained model, whether any of the second plurality of component waveforms comprises an anomaly. For each indicated waveform, the indicated waveform is reviewed and, for each reviewed waveform not comprising an anomaly, the waveform is labelled.