Online Quality Prediction Research Articles

Enhancing incremental deep learning for FCCU end-point quality prediction

In the refining process, Fluid Catalytic Cracking Unit (FCCU) end-point quality prediction plays an important role in real-time quality monitoring, optimization and control. Due to the process uncertainties (such as changes of raw materials, abrasion of mechanic components, catalyst deactivation, changes in the external environment, etc), data distribution of process data in FCCU is time-varying, which leads to accuracy degradation of the quality prediction model. Therefore, to deal with the time varying characteristic of process data and avoid accuracy degradation of the quality prediction model, real-time processing should be considered. In this paper, an enhancing incremental deep learning approach is proposed for the online quality prediction of the absorption-stabilization system in FCCU. First, the offline model is built by Stacked Auto Encoder-Deep Neural Network (SAE-DNN). To determine whether the data distribution has changed and model modification is needed, a concept drift detector is proposed for the regression problem by defining an error bound in the Statistical Test of Equal Proportions (STEPD). If the model modification is needed, then the top layer of the offline SAE-DNN model is expanded by Random Vector Functional Link (RVFL) structure, and the parameters in the expansion layer is dynamically assigned by the new coming data with the group lasso regularization and the L2 regularization. The proposed approach is validated by predicting the Saturated Vapor Pressure (SVP) of stabilized gasoline in the FCCU. The experimental results show that the proposed approach can deal with the time-varying characteristic of process data and avoid accuracy degradation under process uncertainties.

Phase division and transition modeling based on the dominant phase identification for multiphase batch process quality prediction

Batch processes are carried out from one steady phase to another one, which may have multiphase and transitions. Modeling in transitions besides in the steady phases should also be taken into consideration for quality prediction. In this paper, a quality prediction strategy is proposed for multiphase batch processes. First, a new repeatability factor is introduced to divide batch process into different steady phases and transitions. Then, the different local cumulative models that considered the cumulative effect of process variables on quality are established for steady phases and transitions. Compared with the reported modeling methods in transitions, a novel just-in-time model can be established based on the dominant phase identification. The proposed method can not only consider the dynamic characteristic in the transition but also improve the accuracy and the efficiency of transitional models. Finally, online quality prediction is performed by accumulating the prediction results from different phases and transitions. The effectiveness of the proposed method is demonstrated by penicillin fermentation process.

Optimization of time‐variable‐parameter model for data‐based soft sensor of industrial debutanizer

SummaryThe enhancement of modern process control methods has caused the popularity of soft sensors in online quality prediction. It is significant to consider the reduction of model complexity, the performance increment, and decrement of input variables in soft sensor design, simultaneously. The aim of this paper is designing and applying a new data‐based soft sensor with minimum input variables for the enhancement of product quality estimation. Time‐varying‐parameter model by employing the Kalman filter and fixed interval smoothing algorithms has been developed to determine the dynamic transfer function and parameters setting based on time. A novel hybrid method with a dynamic autoregressive exogenous variable model and genetic algorithm has been presented for both state identification and parameter prediction. The combinatorial optimization problem has constructed based on a selection of input variables and an evaluation of Akaike information criterion as a fitness function. An industrial debutanizer column has been used for soft sensor performance validation. The result has indicated that the final soft sensor model in comparison to other presented soft sensing methods for this case has less complexity, fewer input variables, more robust and higher predictive performance. Due to fewer input variables, rapid convergence, and low complexity of this model, it can be efficient in industrial processes control, time‐saving, and improvement of quality prediction.

Runtime Quality Prediction for Web Services via Multivariate Long Short-Term Memory

Online quality prediction helps to identify the web service quality degradation in the near future. While historical web service usage data are used for online prediction in preventive maintenance, the similarities in the usage data from multiple users invoking the same web service are ignored. To improve the service quality prediction accuracy, a multivariate time series model is built considering multiple user invocation processes. After analysing the cross-correlation and similarity of the historical web service quality data from different users, the time series model is estimated using the multivariate LSTM network and used to predict the quality data for the next few time series points. Experiments were conducted to compare the multivariate methods with the univariate methods. The results showed that the multivariate LSTM model outperformed the univariate models in both MAE and RMSE and achieved the best performance in most test cases, which proved the efficiency of our method.

Quality Indexes Design for Online Monitoring Polymer Injection Molding

Quality control is a crucial issue in the injection molding process with target of obtaining a high yield rate and reducing production cost. Consequently, effective methods for monitoring the injection conditions (e.g., pressure, velocity, and temperature) in real-time and adjusting these conditions adaptively as required to ensure a consistent part quality are essential. This study proposes a quality index based on the clamping force increment during the injection cycle, as determined by four strain gauges attached to the tie bars of the injection molding machine. Also, various quality indexes for online quality monitoring and prediction purposes based on the pressure, viscosity, and energy features extracted from the pressure profiles obtained at the load cell, nozzle, and molding cavity, respectively, are compared. The feasibility of the proposed quality indexes is investigated experimentally for various settings of the barrel temperature, back pressure, and rotational speed of the plasticizing screw. It is shown that all of the quality indexes are correlated with the injection-molded quality and hence provide a feasible basis for the realization of an on-line quality monitoring and control system. Particularly, the tie-bar elongation quality index requires no modification or invasion of the injection molding system or cavity and hence provides a particularly attractive solution for monitoring and controlling the part quality.

Online prediction of quality‐related variables for batch processes using a sequential phase partition method

Batch processes inherently have multiple operation phases; different phases exhibit different characteristics. Hence, it is reasonable to partition the process into phases and build sub‐phase models for online quality prediction. To this end, a sequential phase partition method based on the information increment is proposed. To address the multiphase behaviours in batch processes, this work utilizes a new information increment index to capture the dynamic characteristics of batch processes along a time direction and divides the process into sub‐phases. Next, phase‐based multiway partial least squares (MPLS) models are built to model within‐phase characteristics and predict the quality‐related variables online. Information increment is able to exploit the process evolution by focusing on the changing variable correlations derived from two adjacent extend time slice. It directly utilizes the available process measurements of successful history batch processes without data transformation or dimensionality reduction. The method is sequential and can overcome the limits of some phase partition methods that may divide the samples with discontinuous time sequence but similar characteristics into the same phase. In addition, the information increment is capable of reflecting the change of the process intuitively with high computation efficiency. Advantages of the proposed method are illustrated by two case studies, a penicillin simulation platform and an industrial application of Escherichia coli (E. coli) fermentation, respectively.

Recursive partitioning algorithm in water quality prediction

Water quality prediction is the greatest thing in Health and Family Welfare Department. Potable water consumed by citizens can be safely used for drinking, cooking and washing. The contaminated water affects the public health and mostly affects the infants or utero health. Online water quality prediction tool is very useful to every people especially every mother. In this paper, water quality is predicted by giving the longitude and latitude of one location. Compared with linear model and support vector machine, recursive partitioning method is more practical, reasonable and acceptable. Decision trees have less error rates, high data capabilities and good data quality. The predicting results have shown that this recursive partitioning method could analyze and predict the water quality parameters deeply and precisely. The method proposed in this paper is more accurate and could be the decision support for every citizen and every mother. Online aqua prediction tool reduces the needs of man power and equipments.

On-Line Surface Roughness Prediction by Using a Probabilistic Approach for End-Milling

An effective and efficient methodology is proposed to predict surface roughness by online monitoring of surface quality using accelerometer signals. A probabilistic approach, Monte Carlo Simulation, was researched and developed to create an automated tool for on-line prediction of surface quality. Data from 3-axis vibration (Vx, Vy, Vz) signals were used to predict on-line surface roughness. According to an experimental design with four cutting parameters (Cutting speed (Vc), Feed per teeth (Sz), Dept of cut (Dc), Width of cut (Wc)), three-axis vibration signals were used to combine data into a probabilistic model for development of an on-line surface roughness prediction system. Once the probability model was established by using a data set consisting of 71 experiments, the model was tested for 10 different cutting conditions. The probability model shows that the results have convergence values that are close to each other, by as high as 96.37%.

Quality prediction for multi-grade processes by just-in-time latent variable modeling with integration of common and special features

To cope with the difficulty of on-line quality prediction for multi-grade processes widely operated in process industries, a just-in-time latent variable modeling method is proposed based on extracting the common and special features of multi-grade processes. Considering the complicated nonlinear characteristics of multi-grade processes encountered in engineering applications, a just-in-time learning (JITL) strategy is developed to choose the relevant samples from different grades with respect to the query sample. A novel common feature extraction algorithm is proposed to determine the common directions shared by different grades of processes. After extracting the common features, a partial least-squares modeling algorithm is used to extract the special directions for each grade, respectively. Hence, product quality prediction can be simply conducted by integrating the common and special parts of each grade for model building in terms of a JITL strategy. A numerical case and an industrial polyethylene process are used to demonstrate the effectiveness and advantage of the proposed method.

Soft Sensor Modeling Based on Multi-State-Dependent Parameter Models and Application for Quality Monitoring in Industrial Sulfur Recovery Process

Soft sensors have gained wide popularity in the industrial processes for online quality prediction in the recent years. In the case of online deployment, it is important to incorporate fewer input variables to improve the performance of the soft sensor. Therefore, the goal of this paper is to present an approach for the development of more efficient and less complex soft sensors in order to maximize the accuracy as well as to minimize the number of input soft sensing variables. The approach is based on multi-state-dependent parameter (MSDP) models, in which model parameters are estimated in a multivariable state space employing the Kalman filter and fixed interval smoothing algorithms. The proposed MSDP-based soft sensor is applied to an industrial sulfur recovery unit (SRU) in order to predict of H2S and SO2 concentrations. The model is consequently compared with the other soft sensing techniques, which are based on the same benchmark data set of the case study. The prediction results show that the designed MSDP-based soft sensors are more robust and exhibit higher predictive performance than other presented soft sensing methods based on the root mean square errors and Pearson correlation coefficient criterions while using fewer input variables.

RLS-based quality control method for reassembly under uncertainty

The uncertainty recognition, quantification, and optimization of reassembly (remanufacturing assembly) are crucial to improve the precision, quality, and stability of the assembly of remanufactured products. In this paper, we summarize the uncertainties based on the analysis of reassembly features. An uncertainty measure model using information-entropy theory is then established. This model characterizes the uncertainty of remanufactured parts, reused parts, and reassembly quantitatively. An online quality prediction method for reassembly process is then proposed. The method can effectively estimate the quality of remanufactured products using recursive least squares algorithm, which combines real-time data with tremendous amount of history data from the reassembly process. This method can also maximize the use of tremendous amount of history data and improve the accuracy of reassembly by mathematical statistics and examples. Moreover, it can reduce the cost of rework and improve the quality of remanufactured products. Results show that the proposed method achieved good application results with high accuracy and computation efficiency.

Trajectory‐based phase partition and multiphase multilinear models for monitoring and quality prediction of multiphase batch processes

AbstractNew process monitoring and quality prediction methods are proposed for the batch process with multiple operation phases. First, a trajectory‐based phase partition method is developed to divide a batch process into different operation phases by clustering the time slices of reference batches using the warped K‐means algorithm. Multilinear modeling methods, eg, parallel factor analysis and N‐way partial least squares (NPLS), are then used to model the 3‐way batch data in each operation phase. An online process monitoring method is proposed based on the multiphase parallel factor analysis models. An online quality prediction method is developed based on 2‐level quality prediction models, consisting of the first‐level multiphase multiway partial least squares models and the second‐level multiphase NPLS models. The first‐level multiway partial least squares models carry out real‐time quality prediction at each sampling time in different operation phases. At the end of each operation phase, the second‐level multiphase NPLS model is used to compute a more accurate quality prediction by taking into account the phase accumulative effect on the final produce quality. The implementation, effectiveness, and advantages of the proposed methods are illustrated with a case study on a penicillin fermentation process.

Hurdle Modeling for Defect Data with Excess Zeros in Steel Manufacturing Process

The modern steel industry aims to produce high-quality products with higher product yield, lower costs, and lower energy consumption to meet market demands. To accomplish these goals, it is necessary to reduce or eliminate product defects. However, the relationship of operating conditions to the defect formation is not fully understood. There is increasing interest in developing models to monitor the quality and predict the number of defects in real time. Modeling and analyzing the defect count data is a very challenging problem because the defect count data exhibit the unique characteristics of non-negative integers, overdispersion, high skewed distribution, and excess zeros. To explicitly account for these unique characteristics, the present work develops an on-line quality monitoring and prediction system based on the hurdle regression model. The basic idea of the hurdle model is that a binomial model governs the binary outcome of the dependent variable being zero or positive. If the dependent variable takes a positive value, ”hurdle is crossed”, and the conditional distribution of the positives can be modeled by a zero-truncated Poisson or negative binomial (NB) model. Compared to Poisson and NB models, the hurdle model is not only suitable for modeling discrete and non-negative integer data, but also sufficient for handling both overdispersion and excess zeros data. The effectiveness of the hurdle model was verified through its application to the real defect data of a steelmaking plant. The results have demonstrated that the hurdle NB model is superior to the Poisson, NB, hurdle Poisson, and PLS models in the prediction performance.

Novel common and special feature extraction method for modeling multi-grade processes

In the processing industries, operating conditions often change to meet the requirements of the market and customers. To cope with the difficulty of on-line quality prediction for such multi-grade processes widely operated in process industries, a novel common and special feature extraction method is proposed for modeling multi-grade processes. A common feature extraction algorithm is proposed to determine the common directions shared by different grades of these processes. After extracting the common features, a partial least-squares modelling algorithm is used to extract the special directions for each grade, respectively. Hence, product quality prediction can be simply conducted by integrating the common and special parts of each grade for model building. A numerical case and an industrial polyethylene process are used to demonstrate the effectiveness and advantage of the proposed method.

Self-adaptive evolving forecast models with incremental PLS space updating for on-line prediction of micro-fluidic chip quality

An important predictive maintenance task in modern production systems is to predict the quality of products in order to be able to intervene at an early stage to avoid faults and waste. Here, we address the prediction of the most important quality criteria in micro-fluidics chips: the flatness and critical size of the chips (in the form of RMSE values) and several transmission characteristics. Due to semi-manual inspection, these quality criteria are typically measured only intermittently. This leads to a high-dimensional batch process modeling problem with the goal of predicting chip quality based on the trends in these process values (time series). We apply time-series based transformation for dimension reduction to the lagged time-series space using of partial least squares (PLS), and combine this with a generalized form of Takagi–Sugeno(TS) fuzzy systems to obtain a non-linear PLS forecast model (termed as PLS-fuzzy). The rule consequent functions are robustly estimated by a weighted regularization scheme based on the idea of the elastic net approach. To address particular system dynamics over time, we propose dynamic updating of the non-linear PLS-fuzzy models using new on-line time-series data, with the options 1.) adapt and evolve the rule base on the fly, 2.) smoothly down-weight older samples to increase flexibility of the fuzzy models, and 3.) update the PLS space by incrementally adapting the loading vectors, where processing is achieved in a single-pass stream mining manner. We call our method IPLS-GEFS (incremental PLS combined with generalized evolving fuzzy systems). We applied our predictive modeling approach to data from on-line microfluidic chip production over a time period of about 6 months (July to December 2016). The results show that there is significant non-linearity in the predictive modeling problem, as the non-linear PLS-fuzzy modeling approach significantly outperformed classical PLS for most of the targets (quality criteria). Furthermore, it is important to update the models on the fly with incremental updating of the PLS space and/or with down-weighting older samples, as this significantly decreased the accumulated error trends of the prediction models compared to conventional updating. Reliable predictions of flatness quality (with around 10% error) and of RMSE values and transmissions (with around 15% errors) can be achieved with prediction horizons of up to 4 to 5 h into the future.

Quality by control: Towards model predictive control of mammalian cell culture bioprocesses.

The industrial production of complex biopharmaceuticals using recombinant mammalian cell lines is still mainly built on a quality by testing approach, which is represented by fixed process conditions and extensive testing of the end-product. In 2004 the FDA launched the process analytical technology initiative, aiming to guide the industry towards advanced process monitoring and better understanding of how critical process parameters affect the critical quality attributes. Implementation of process analytical technology into the bio-production process enables moving from the quality by testing to a more flexible quality by design approach. The application of advanced sensor systems in combination with mathematical modelling techniques offers enhanced process understanding, allows on-line prediction of critical quality attributes and subsequently real-time product quality control. In this review opportunities and unsolved issues on the road to a successful quality by design and dynamic control implementation are discussed. A major focus is directed on the preconditions for the application of model predictive control for mammalian cell culture bioprocesses. Design of experiments providing information about the process dynamics upon parameter change, dynamic process models, on-line process state predictions and powerful software environments seem to be a prerequisite for quality by control realization.

Multi-Phase Support Vector Regression Soft Sensor for Online Product Quality Prediction in Glutamate Fermentation Process

Glutamate fermentation is inherently nonlinear, multi-phase and an aerobic fermentation process. As long measurement delays and expensive apparatus cost, on-line measurement of the product concentration is not necessarily available. The present fermentation process monitoring and quality prediction involve manual interpretation of highly informative, however, the concentrations of substrates, biomass and products are only low frequency off-line measurements. In this paper, we propose a novel Multi-Phase Support Vector Regression (MPSVR) based soft sensor model for online quality prediction of glutamate concentration. The glutamate fermentation process can be divided into a sequence of five phases by detecting the trend variation events (also termed as singular points or inflection point) of online measured O2 in the exhaust gas, the Inflection Point (IP) are easily identified through combining Moving Window (MW) with Pearson Correlation Coefficient (PCC). For each estimation phase, SVR soft sensor model are constructed and their performance is evaluated against fermentation data in a 5 L fermenter. The efficiency of the proposed soft sensor model for online product quality prediction has been demonstrated to be superior compared to that of reported techniques in a 5 L glutamate fermentation process.

On-line quality prediction of batch processes using a new kernel multiway partial least squares method

The conventional data-driven soft sensor methods such as multiway partial least squares have been encountering nonlinear problems in predictions of batch processes, and kernel methods have been used to deal with these problems. In this work, a new data-driven soft sensor method is proposed by developing a Reduced Dual Kernel multiway partial least squares algorithm. First, the number of kernel vectors is reduced by the feature vector selection method. Then, by projecting both input data and the output data into two reduced kernel spaces, dual kernel matrices are established. These two matrices can be used to build PLS models. Finally, the predicted data in the kernel space can be reversely projected onto its original space during online prediction. Comparisons were made among the proposed method and some pervious algorithms through a numerical example and an Escherichia coli fermentation batch process.

Sequential and Orthogonalized Partial Least-Squares Model Based Real-Time Final Quality Control Strategy for Batch Processes

In this work, the problem of driving a batch process to a desired final product quality using data-driven model based midcourse correction (MCC) is described. Specifically, we adapt a sequential and orthogonalized partial least-squares (SO-PLS) method to calibrate the inferential quality model, which takes into account the serial nature of the input batch data matrices and could retain the reliable information as much as possible when it is used to perform online quality prediction. Since the process variable trajectories that are necessary to predict the final quality are incomplete at a certain decision point, known data regression (KDR) is used to estimate the future trajectories, and the causal relationship of the initial conditions and the future candidate manipulated variables in determining the future process variable trajectories is also considered. Finally, taking the advantage of the latent variable model, the indicators that consider only the degrees of freedom are employed as hard constraints ...

Soft sensor development for online quality prediction of industrial batch rubber mixing process using ensemble just-in-time Gaussian process regression models

Rubber mixing is a nonlinear batch process that lasts for very a short time (ca. 2–5min). However, the lack of online sensors for quality variable (e.g., the Mooney viscosity) has become a main obstacle of controlling rubber mixing accurately, automatically and optimally. This paper proposes a novel soft sensing method based on Gaussian process regression (GPR) models fortified with both ensemble learning and just-in-time (JIT) learning, which ensures precision and robustness at the same time. More specifically, this method first builds multiple input variable sets from random local datasets, then uses the obtained input variable sets to establish local models and send them to ensemble learning with Bayesian inference and finite mixture mechanism before making the final prediction output. The superiority of the proposed method is demonstrated using an industrial rubber mixing process.