Time-Series Segmentation for Maintenance-Aware Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In semiconductor manufacturing, existing methods struggle to quickly and accurately detect anomalies in equipment data due to noise and changes caused by preventive maintenance, making it difficult to take timely corrective action and maintain product yield and quality.
Innovation Solution
The method involves data cleaning to remove outliers, segmentation of the data based on step changes, and statistical modeling to identify anomalies, using techniques such as change-point detection and rolling window algorithms, followed by linear fitting to determine slope and intercept comparisons for anomaly detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If statistical methods are used to monitor equipment data, then data can be evaluated to define key indicators, but anomalies are not always effectively revealed due to noise and changes from preventive maintenance
Solution Approach 1:
The patent segments the time-series data into multiple segments based on detected change points, which represent transitions in equipment state or maintenance events. This segmentation allows the system to analyze each segment separately with appropriate statistical models, improving anomaly detection accuracy by avoiding contamination from unrelated data variations and maintenance-induced changes.
Solution Approach 2:
The patent performs preliminary data cleaning to remove outliers and automatically detects change points before conducting anomaly detection. By preprocessing the data to eliminate obvious anomalies and segmenting at change points, the system prepares the data in advance, making the subsequent anomaly detection more reliable and less susceptible to false positives from maintenance activities.
2Reliability
If preventive maintenance activities are performed on equipment, then equipment reliability is improved, but sudden changes in data values or trends occur making anomaly identification difficult
Solution Approach 1:
The patent automatically detects change points in the time-series data that correspond to maintenance events or state transitions, and segments the data at these points. This allows the system to distinguish between expected changes due to maintenance and actual anomalies, making anomaly identification easier without compromising equipment reliability improvements from maintenance.
Solution Approach 2:
The system continuously monitors equipment data and provides feedback by detecting change points and segmenting data accordingly. This feedback mechanism allows the system to adapt to maintenance events dynamically, adjusting its analysis approach to maintain effective anomaly detection despite data changes from maintenance activities.
3Measurement precision
If data is cleaned to remove outliers and segmented, then anomaly detection accuracy is improved, but processing complexity increases
Solution Approach 1:
The patent applies segmentation based on automatically detected change points, which divides the data into manageable segments that can be analyzed independently. While this improves anomaly detection accuracy by focusing analysis on relevant data portions, the automated change point detection and segmentation process adds processing complexity that requires sophisticated algorithms.
Solution Approach 2:
The patent performs preliminary data cleaning and outlier removal before anomaly detection. This preliminary action improves detection accuracy by eliminating noisy data, but it also increases processing complexity by adding preprocessing steps that must be executed before the main analysis.
Data Source
AI summary
Detection of data anomalies resulting from maintenance activities on semiconductor processing equipment. Time-series representation of the key indicators for equipment performance is cleaned then segmented according to sharp breaks in the data. The cleaned and segmented data is modeled, for example, by determining a linear fit for each segment. The slope and intercept of each modeled segment linear fit are compared and evaluated to identify anomalies in the data.