Modeling And Fault Monitoring Of Batch Process Using Multiphase AR-PCA

Batch process is the main production method in the biopharmaceutical,fine chemical,food and beverage production industries.However,due to its intermittent characteristic,there are periodic mass production,dynamic state of material status and operating parameters,and high process control requirements in batch process.The fermentation process is a typical intermittent process.The fermentation process is related to economic development and improvement of the people's living standards.It is one of the seven strategic emerging industries established by the State Council and it will play an important role in the economic integration of Beijing,Tianjin and Hebei.Effect.The project focuses on the issues of batch unequal length characteristics,dynamic characteristics,and multi-phase characteristics of the biological fermentation process,the project studied the problems existing in the previous methods of monitoring and established a high-efficiency and high-precision process monitoring model to reduce the false alarm rate and leak alarm rate of monitoring.To ensure the safety of operations and timely capture the changes in the detection variables during the fermentation process.If any monitoring failures are found,the employees are notified in a timely manner.The staff adjusts the fermentation environment or suspends production to maximize the product quality,stabilize production,or reduce losses.This will reduce energy consumption and waste of resources.Once the research results are popularized,it will greatly increase the safety of production during the fermentation process,reduce accidents and waste of resources,and create greater economic and social benefits.The main work of this article is as follows:(1)The multi-constraint dynamic time warping is proposes to solve the problem of synchronization problem between batches.In view of the inherent batch inequality problem as well as overcoming the problems such as wasted data,distorted original process variables autocorrelation and cross-correlation associated with traditional methods when solving batch synchronization problems,multi-constraint dynamic time warping(DTW)method was proposed,it dynamically solves batch-to-batch problems by dynamically matching the mid-point and point patterns of the track.(2)An adaptive weighted particle swarm optimization based on population diversity is proposed to optimize the affine propagation clustering phase method.For the multi-stage characteristics of the batch process,the advantages and disadvantages of the phase division method based on the clustering algorithm are analyzed.The particle swarm optimization(PSO)algorithm based on adaptive inertial weights based on population diversity was used to improve the Affinity Propagation(AP)clustering and guide the phase division.(3)Multi-stage AR-PCA is proposed for monitoring of fermentation process.An autoregressive Principal Component Analysis(AR-ARC)was constructed for the data samples of the multistage fermentation process of AP diversity based on Population Diversity-based Particle Swarm Optimization(PDPSO).The PCA model was used to eliminate the autocorrelation and cross-correlation between the dynamics of each stage and the variables.A Principal Component Analysis(PCA)model was built for the monitoring of the fermentation process for the residual matrix of the AR model.(4)Field trials of E.coli fermentationThe proposed method is ultimately applied to practical production.This article relies on the rationality and effectiveness of the research method of E.coli fermentation experimental testing.The multi-stage batch process modeling method proposed in this paper is applied to the field experiment of E.coli fermentation process.The results show that this method can greatly improve the monitoring effect of the production process.When an abnormality occurs in the production process,the alarm is promptly reported and the production efficiency is improved.