Parameter Fault Detection Using Moving-Average Sequence Deviation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In manufacturing processes, especially in semiconductor production, it is challenging to detect behavioral deviations in parameter variables due to varying time sequence lengths and timing relationships among multiple targets, making it difficult to establish a standard sequence for normal conditions, particularly when numerical variations are large and reasonable limit values are hard to set.
Innovation Solution
A fault detection method involving moving average operations and dynamic time warping (DTW) to align and normalize sequences, followed by a DTW-based barycenter average (DBA) operation to establish a standard sequence, and calculating difference sequences to determine abnormal conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional fault detection methods are used to monitor parameter variables in manufacturing processes, then the detection of obvious abnormal conditions is possible, but the detection of behavior deviation and discontinuous conditions becomes difficult when time sequence lengths vary and numerical variations are large
Solution Approach 1:
The patent applies preliminary action by performing multiple moving average operations on the target sequence before comparing it with the standard sequence. Specifically, it calculates a first moving average sequence with a first window length, then calculates a second moving average sequence with a second window length, and finally computes a difference sequence from these averages. This preliminary processing reduces numerical variations and aligns time sequences, making the subsequent fault detection more accurate and the establishment of standard sequences more manageable.
2Reliability
If the time sequences of multiple targets are analyzed directly without preprocessing, then the original data is preserved, but the varying lengths and timing relationships make it difficult to establish a standard sequence for normal conditions
Solution Approach 1:
The patent applies parameter changes by transforming the original time sequence data through multiple moving average operations with different window lengths. The first moving average operation uses a first window length to smooth the data, while the second moving average operation uses a second window length to further process the smoothed data. This transformation changes the parameters of the time sequences, reducing numerical variations and making them more suitable for establishing a reliable standard sequence.
3Ease of manufacture
If moving average operations are performed to reduce numerical variation, then the establishment of standard sequence becomes easier, but the detection of behavior deviation may be affected by the smoothing effect
Solution Approach 1:
The patent applies partial or excessive action by performing multiple moving average operations with different window lengths rather than a single operation. It calculates a first moving average sequence with a first window length and a second moving average sequence with a second window length, then computes the difference between them. This excessive processing with multiple averages ensures that the standard sequence establishment is sufficiently simplified while the difference operation preserves the ability to detect behavior deviations.
Data Source
AI summary
A fault detection method, includes the following steps. A target sequence is received, the target sequence includes several data. A first moving average operation is performed on the target sequence to establish a first moving average sequence. A second moving average operation is performed on the target sequence to establish a second moving average sequence. A difference operation between the first moving average sequence and the second moving average sequence is performed to obtain a difference sequence, the difference sequence includes several difference values. An upper limit value is set. When one of the difference values is greater than the upper limit value, the target sequence is determines as abnormal.