Research On Hardness Prediction And Fault Diagnosis Of Steel Strip In Continuous Annealing Process

Continuous annealing process is an important production procedure of cold rolling mill for iron and steel enterprises,and the hardness of strip after annealing is the core index of product quality and production.Since the online detection of strip hardness cannot be achieved,the hardness detection is usually carried out on the head and tail sections cut from a strip after annealing.Such an offline detection method inevitably has a large time delay,which often causes large fluctuation of hardness or even quality problems such as hardness does not meet the requirement.In addition,the continuous annealing process consists of several production stages such as heating and cooling,which makes the whole process very complex.When a unit fails the fault diagnosis is often based on the workers' experience which cannot achieve timely warning and rapid diagnosis of failure.The two problems have seriously affected the economic benefits of cold rolling mill.Therefore,this dissertation is focused on the online prediction method of strip hardness for continuous annealing production process,as well as the operation monitoring and fault diagnosis for this process,so as to help the cold rolling mill to improve product quality and guarantee the continuous and stable operation of the continuous annealing production process.The detailed research components are as follows:(1)For the data sampling and pre-processing,the collected process data are firstly pre-processed through combining coil number and sampling time so as to construct a complete sample.Subsequently,a two-layer clustering method is proposed to eliminate the sample data with gross error.(2)For the strip hardness modeling of continuous annealing process,the principle component analysis(PCA)is used to reduce the dimension of high dimensional sample data so as to reduce the modeling complexity.In view of the limitation of the generalization ability of a single learning machine,the on-line ensemble learning modeling method based on the least square support vector machine is proposed.(3)For the construction of the single learning machine in the ensemble learning,the parameter determination of each learning machine is viewed as an optimization problem.And for this problem,an improved self-adaptive genetic algorithm based on multiple crossover operators is proposed.Computational results based on both benchmark test problems and practical production data illustrate the efficiency of the proposed algorithm and modeling method.(4)For the process monitoring and fault diagnosis of continuous annealing,the kernel PCA method based on radial basis function is adopted due to the limitations of classical PCA for non-linear problems.Based on the control limit of T2 and SPE obtained from the kernel PCA,whenever the statistical quantity exceeds the corresponding limit,the contribution of each process variable is used to determine the fault source.Through this way,the online monitoring and fault diagnosis of continuous annealing production process can be achieved.