A Fault Monitoring Method for MKECA Fermentation Process Based on Extended Kernel Entropy Load Matrix

Abstract

The invention relates to a fault monitoring method of MKECA fermentation process based on an extended kernel entropy load matrix, which belongs to the technical field of fault monitoring. Including "offline modeling" and "online monitoring". "Offline modeling" step: firstly, the collected 3D data is subjected to dimensionality reduction and standardization processing, and the principal component information of the data is analyzed by using the nuclear entropy component; then the time is extended to the nuclear entropy load matrix to generate the nuclear entropy expansion load matrix, and Calculate the similarity between the kernel entropy expansion load matrices; finally, use the fuzzy-C-means to divide it into stages, establish a KECA monitoring model, and calculate statistics and corresponding control limits. "Online monitoring" step: standardize the collected data, calculate statistics and control limits; compare with the offline control limits, if the limit is not exceeded, the production process is normal, otherwise, the production process is faulty. The invention solves the problem of misclassification of jumping points, makes the divided stages more accurate, and reduces false positive and false negative rates.

Description

technical field [0001] The invention belongs to the technical field of fault monitoring, and relates to a data-driven intermittent process online fault monitoring technology, in particular to a method for stage division of a multi-stage intermittent process. Background technique [0002] In recent years, the batch production method is gradually surpassing the continuous production method to become the mainstream of the market, especially in the fields of medicine and biopharmaceuticals. The multi-stage characteristic is an inherent characteristic of the batch production process. There are different key process variables and control objectives in different stages. It is difficult to obtain satisfactory results by directly using the existing multivariate statistical process monitoring (Multivariate Statistical Process Monitoring, MSPM) method. A very typical phenomenon is that after the current model continues to run for a period of time, there will be a large number of false ...

AI Technical Summary

Problems solved by technology

Most of the above methods obtain the load matrix through dimension reduction processing, and then use the variation direction or variation amplitude of the feature matrix to model, ignoring the existence of unknown noise in the actual production process, resulting in low accuracy and sensitivity of the built model, resulting in a large number of Missed and false alarms

Therefore, fully considering the existence of a large number of unknown noises and singular values ​​in actual production, the traditional method of only relying on the load matrix to establish a monitoring model has certain limitations, resulting in a large number of false alarms, missed alarms, and even failures during online real-time monitoring. There is a certain lag in the alarm time

Images