Quality Prediction And Parameter Design For Complex Industrial Process Based On Computational Intelligence

Generally,complex industrial process refers to a precision product manufacturing process with high added-value through multiple operating stages.In addition,it also can be defined as the basic material or related product production process with complex and implicit physico-chemical reaction.For this industrial process,the control capability of production process quality plays an important role in whether the core industry can produce product with desired quality,improve completive strength and make the country strong through quality development.Theoretically,the recent updating and upgrading of production technology and equipment present more chance for data acquisition and quality control.However,the limited utilization level on production management technique hinders the advantage of hardware.Most of them still use process monitoring and process improvement based on sampling or experimental data to,which is hard to effectively and actively predict final quality condition and derive the adjustment solution for operating parameter with collected production data.Additionally,the existing research work in terms of data-driven quality prediction and parameter design methods still have great research room on combination between computational intelligence technique and features of complex industrial process.Furthermore,the quality index prediction and operating parameter optimization are the kernel module of intelligent optimal manufacturing.Thus,after deeply analyzing the nature of complex industrial process,the novel quality prediction and parameter design method are proposed based on industrial observational data and computational intelligence techniques in this work to alleviate the weakness in existing research.The main research content is summarized as follows:(1)The quality index prediction method is presented based on semi-supervised deep learning techniques.First,the information hidden in considerable unlabeled data is learned by Stacked Auto-Encoder network to reconstruct the labeled data(denoted as latent variable dataset).Then,the supervised Bidirectional Long Short-Term network is used to extract the dynamic feature from the latent variables.Finally,the extracted features are set as inputs for fully connected network to extract higher-level feature and establish fitting model with quality index.The results from a real case study indicate that the proposed method outperforms other benchmark prediction models.In addition,the experimental analysis also shows that the model’s performance may have no obvious improvement when the ratio between number of labeled and unlabeled sample data reach a limit.Thus,the proposed forecasting model can be severed as a soft sensor to support balanced and timely operating data.(2)A novel multistage and multi-model strategy driven intelligent parameter design approach is proposed for complex industrial process according to ‘Separately modeling for individual operating stage and then connecting sub-models for whole production process’ principle.Herein,the black-box machine learning method is first applied to construct the correlation model between input variables(including auxiliary parameters and controllable parameters)and quality indexes within operating stage.Then,the intelligent optimization algorithm is used to solve the multi-objective programming model combing with desirability function based multi-objective transferring method.Besides,two different connecting modes between adjacent operating stage are analyzed and compared in this section.The experimental results demonstrate the proposed black-box intelligent parameter design scheme can effectively derive the optimal operating parameters setting value.Additionally,the rational quality specification at various intermediate stages also can be obtained by proposed method,which is extremely important for early abnormality detection.(3)Based on same modeling principle,a new intelligent parameter design model with explicit mathematical equation is presented by integrating multi-Gene Genetic Programming,multi-objective intelligent optimization algorithm with fuzzy multicriteria decision-making methods for multistage and multi-response industrial process.The proposed parameter design method in this section also can alleviate the problems which caused by nature of complex industrial process.The application in illustrating real-time industrial case for proposed method shows the proposed parameter design method has more excellent performance than black-box intelligent models due to the explicit correlation expression and direct multiple objective optimization algorithm.Besides,all the two proposed parameter design schemes with multistage and multimodel strategy in this work are more promising and effective than existing single-model parameter design methods.The abovementioned research tries to discuss the possibility that using computational intelligence techniques-driven solution with collected industrial operating data to alleviate the quality problems in complex industrial process under new industry condition,which enriches the research scope and content for quality management and intelligent manufacturing.Meanwhile,our research work is beneficial for engineers to make decision for engineers by presenting detailed and effective solution for early prediction and operating parameter adjustment to obtain desired final product quality.