Batch Synchronization for Accurate Industrial Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection systems in industrial batch analytics often produce high false positive or false negative results due to inconsistent accuracy, leading to unnecessary production stops or unqualified products.
Innovation Solution
The system uses a method that determines an anomaly metric based on T2-statistic and Q-statistic metrics in a Principal Component Analysis (PCA) model, normalizes these metrics, and selects the highest value for comparison, while also employing a machine learning model to identify the best anomaly detection threshold, enabling accurate detection of anomalies in both complete and ongoing batches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional anomaly detection methods are used in industrial batch analytics, then the system can detect anomalies, but it produces high false positive or false negative results due to inconsistent accuracy
Solution Approach 1:
The patent transforms batch data of varying lengths into fixed-length representations by mapping to reference batches. This parameter transformation enables consistent comparison across batches with different durations, resolving the inconsistency in detection accuracy while maintaining reliable anomaly detection performance.
Solution Approach 2:
The patent creates reference batches from historical non-anomalous data and uses them as templates to transform and normalize incoming batch data. By copying the temporal structure of reference batches and mapping new data onto this standardized framework, the system achieves consistent detection accuracy across varying batch lengths.
2Productivity
If batch data of different lengths is analyzed directly, then real-time monitoring is possible, but alignment and comparison between batches become inconsistent
Solution Approach 1:
The patent implements dynamic time warping techniques that allow flexible alignment of batch data with reference batches. This dynamic approach enables real-time monitoring of batches with varying lengths by adaptively mapping temporal points, maintaining both productivity through continuous monitoring and precision through accurate alignment.
Solution Approach 2:
The system transforms the temporal parameter of batch data by resampling and mapping variable-length batches to fixed-length reference representations. This parameter transformation enables direct comparison while preserving real-time monitoring capabilities, as the transformation can be applied continuously to incoming data streams.
3Reliability
If multiple anomaly detection metrics are used, then detection coverage is improved, but system complexity increases
Solution Approach 1:
The patent combines multiple anomaly detection metrics (T2-statistic from PCA and Q-statistic) into a unified detection framework. By merging these complementary metrics and integrating them with the batch alignment methodology, the system achieves comprehensive anomaly detection coverage while managing complexity through a cohesive architectural approach.
Data Source
AI summary
An illustrative method includes a batch analytic system receiving batch data of a batch generated in an industrial process, wherein the batch data includes a set of samples associated with the batch, the batch is complete and has a first batch length, determining a reference batch based on a plurality of non-anomalous batches generated in the industrial process, wherein each non-anomalous batch has a same second batch length, generating a batch representation of the batch based on the batch data of the batch and the reference batch, wherein the batch representation of the batch aligns with the reference batch and has the second batch length associated with the reference batch, and performing an operation using the batch representation of the batch.