Multimodal Process Fault Detection Method Based on Local Neighbor Normalization Matrix

Abstract

Multimodal process fault detection method based on local neighbor normalization matrix, involving industrial process fault detection method, using historical data under normal state as the training set of modeling data, using local neighbor normalization matrix method for multimodal process modeling and Troubleshooting. Firstly, the local weighting algorithm is used to preprocess the unequal-length batch data. Determine the maximum retainable length of unequal-length data in the training samples, and use k-nearest neighbor information to reconstruct the missing data points of unequal-length batches through weighting. Secondly, the main local neighbor normalization matrix is ​​constructed for the training set of equal length, the K-means algorithm is used for mode clustering, and the outlier samples of each mode are eliminated by the local outlier factor method. This method can avoid the loss of information from affecting the modal clustering effect of the multi-modal process, and remove outliers at the same time. By constructing the primary and secondary local neighbor normalization matrix, the fault diagnosis results of the multi-modal intermittent process are more accurate.

Description

technical field [0001] The invention relates to an industrial process fault detection method, in particular to a multi-mode process fault detection method based on a local neighbor normalization matrix. Background technique [0002] Multimodal processes to produce high-value and diverse products are more prevalent in order to meet changes such as market demand and raw materials. Compared with the traditional batch process, the multi-mode batch process is more complicated, and has the characteristics of severe nonlinearity, time-varying and multi-working conditions, which makes the fault diagnosis of the multi-mode batch process more challenging. In recent years, many scholars have analyzed multi-modal industrial processes from different perspectives, and proposed a variety of fault diagnosis methods. These methods do not need to assume that the data obeys a single distribution, and have a good monitoring effect on the chemical process with complex data distribution, and ope...

AI Technical Summary

Problems solved by technology

[0003] Due to the production characteristics of the batch process, the production cycle of different batches is different, and the problem of unequal length batches will inevitably occur

On the other hand, on-site process data inevitably contain different degrees of error, measurement noise and system noise, etc. These problems will bring certain pollution to the data, making the multi-modal production process data produce local outliers

Such outliers will cause the direction of the pivot to shift, and if such data is not analyzed and preprocessed before fault detection, it will affect the accuracy of the final fault diagnosis

How to ensure that the loss of information is avoided, there is currently no method for fault detection of multi-modal unequal-length intermittent processes

Images