Online Quality Prediction Research Articles

Deep quality-related stacked isomorphic autoencoder for batch process quality prediction

Abstract Batch processes play an important role in modern chemical industrial and manufacturing production, while the control of product quality relies largely on online quality prediction. However, the complex nonlinearities of batch process and the dispersion of quality-related features may affect the quality prediction performance. In this paper, a deep quality-related stacked isomorphic autoencoder for batch process quality prediction is proposed. Firstly, the same raw input data is reconstructed layer-by-layer by isomorphic autoencoder and the raw data features are obtained. Secondly, the correlation between the isomorphic representations of each layer and the output is analyzed by maximum information coefficient to construct the relevant loss function and enhance the quality-related information. Thirdly, deep quality-related prediction model is constructed to predict the batch process quality variables. Finally, the effectiveness of the proposed method is verified by applying on penicillin fermentation process.

Supervised probabilistic dynamic-controlled latent-variable model for quality pattern prediction and optimisation

A supervised probabilistic dynamic-controlled latent-variable (SPDCLV) model is proposed for online prediction, as well as real-time optimisation of process quality indicators. Compared to existing probabilistic latent-variable models, the key advantage of the proposed method lies in explicitly modelling the dynamic causality from the manipulated inputs to the quality pattern. This is achieved using a well-designed, dynamic-controlled Bayesian network. Furthermore, the algorithms for expectation-maximisation, forward filtering, and backward smoothing are designed for learning the SPDCLV model. For engineering applications, a framework for pattern-based quality prediction and optimisation is proposed, under which the pattern-filtering and pattern-based soft sensor are explored for online quality prediction. Furthermore, quality optimisation can be realised by directly controlling the pattern to the desired condition. Finally, case studies on both an industrial primary milling circuit and a numerical example illustrate the benefits of the SPDCLV method in that it can fully model the process dynamics, effectively predict and optimise the quality indicators, and monitor the process.

Implementing multiblock techniques in a full-scale plant scenario: On-line prediction of quality parameters in a continuous process for different acrylonitrile butadiene styrene (ABS) products

BackgroundThe study explores the challenges of handling multiblock data of different natures (process and NIR sensors) for on-line quality prediction in a full-scale plant scenario, namely a plant operating in continuous on an industrial scale and producing different grade Acrylonitrile Butadiene Styrene (ABS) products. This environment is an ideal scenario to evaluate the use of multiblock data analysis methods, which can enhance data interpretation, visualization, and predictive performances. In particular, a novel multiblock extension of Locally Weighted PLS has been proposed by the authors, namely Locally Weighted Multiblock Partial Least Squares (LW-MB-PLS). Response-Oriented Sequential Alternation (ROSA) has also been employed to evaluate the diverse block relevance for the prediction of two quality parameters associated with the polymer. Data are split in blocks both according to sensor type and different plant sections, and different models have been built by incremental addition of data blocks to evaluate if early estimation of product quality is feasible. ResultsROSA method showed promising predictive performance for both quality parameters, highlighting the most influential plant sections through the selection of data blocks. The results suggested that both early and late-stage sensors play crucial roles in predicting product quality. A reasonable estimation of quality parameters before production completion has been achieved. On the other hand, the proposed LW-MB-PLS, while comparable in predictive performances, allowed reducing systematic prediction errors for specific products. SignificanceThis study contributes valuable insights for continuous production processes, aiding plant operators and paving the way for advancements in online quality prediction and control. Furthermore, it is implemented as a locally weighted extension of MB-PLS.

Predicting part quality early during an injection molding cycle

Abstract Data-based process monitoring in injection molding plays an important role in compensating disturbances in the process and the associated impairment of part quality. Selecting appropriate features for a successful online quality prediction based on machine learning methods is crucial. Time series such as the injection pressure and injection flow curve are particularly suitable for this purpose. Predicting quality as early as possible during a cycle has many advantages. In this paper it is shown how the recording length of the time series affects the prediction performance when using machine learning algorithms. For this purpose, two successful molding quality prediction algorithms (k Nearest Neighbors and Ridge Regression) are trained with time series of different lengths on extensive data sets. Their prediction performances for part weight and a geometric dimension are evaluated. The evaluations show that recording time series until the end of a cycle is not necessary to obtain good prediction results. These findings indicate that early reliable quality prediction is possible within a cycle, which speeds up prediction, allows timely part handling at the end of the cycle and provides the basis for automated corrective interventions within the same cycle.

Supervised Multi-Layer Conditional Variational Auto-Encoder for Process Modeling and Soft Sensor.

Variational auto-encoders (VAE) have been widely used in process modeling due to the ability of deep feature extraction and noise robustness. However, the construction of a supervised VAE model still faces huge challenges. The data generated by the existing supervised VAE models are unstable and uncontrollable due to random resampling in the latent subspace, meaning the performance of prediction is greatly weakened. In this paper, a new multi-layer conditional variational auto-encoder (M-CVAE) is constructed by injecting label information into the latent subspace to control the output data generated towards the direction of the actual value. Furthermore, the label information is also used as the input with process variables in order to strengthen the correlation between input and output. Finally, a neural network layer is embedded in the encoder of the model to achieve online quality prediction. The superiority and effectiveness of the proposed method are demonstrated by two real industrial process cases that are compared with other methods.

Real-time rubber quality model based on CNN-LSTM deep learning theory

Predicting the quality of a rubber compound is a necessity for intelligent mixing. However, the conventional quality prediction based on the power curve has low precision and significant dispersion, making it challenging to satisfy the need for real-time rubber mixing quality inspection. In this research, the vibration signals were collected for the first time as an input feature of the mixing quality prediction model, with carbon black dispersion is utilized as a quality index. On the basis of the theory of deep learning, the online quality prediction model of mixing was constructed using a variety of featured extraction methods and neural network structures, and the models were compared and tested. After 5 experiments, the average root mean square error (RMSE) of convolutional neural network-long short-term memory network (CNN-LSTM) model is 0.1160, which is 21.46% higher than that of LSTM model after time-domain featured extraction, demonstrating the efficacy and superiority of CNN-LSTM end-to-end model. This study is essential for the progression and breakthrough of the real-time mixing quality optimization issue.

Online Prediction of Molded Part Quality in the Injection Molding Process Using High-Resolution Time Series.

Process-data-supported process monitoring in injection molding plays an important role in compensating for disturbances in the process. Until now, scalar process data from machine controls have been used to predict part quality. In this paper, we investigated the feasibility of incorporating time series of sensor measurements directly as features for machine learning models, as a suitable method of improving the online prediction of part quality. We present a comparison of several state-of-the-art algorithms, using extensive and realistic data sets. Our comparison demonstrates that time series data allow significantly better predictions of part quality than scalar data alone. In future studies, and in production-use cases, such time series should be taken into account in online quality prediction for injection molding.

Towards a Deep Learning-based Online Quality Prediction System for Welding Processes

The digitization of manufacturing processes enables promising applications for machine learning-assisted quality assurance. A widely used manufacturing process that can strongly benefit from data-driven solutions is gas metal arc welding (GMAW). The welding process is characterized by complex cause-effect relationships between material properties, process conditions and weld quality. In non-laboratory environments with frequently changing process parameters, accurate determination of weld quality by destructive testing is economically unfeasible. Deep learning offers the potential to identify the relationships in available process data and predict the weld quality from process observations. In this paper, we present a concept for a deep learning based predictive quality system in GMAW. At its core, the concept involves a pipeline consisting of four major phases: collection and management of multi-sensor data (e.g. current and voltage), real-time processing and feature engineering of the time series data by means of autoencoders, training and deployment of suitable recurrent deep learning models for quality predictions, and model evolutions under changing process conditions using continual learning. The concept provides the foundation for future research activities in which we will realize an online predictive quality system for running production.

Comparison of online quality prediction models of kiwifruit at different conveying speeds

Comparison of online quality prediction models of kiwifruit at different conveying speeds

Quality analysis for condition monitoring of a manufacturing process integrating acoustic processing based on three distinct ML algorithms

Abstract Over the years, it has been proven that all machines that run continuously or even for a short period of time experiences breakdowns, reduced efficiency and damage to the parts. Considering one such manufacturing process is welding process where strip welding is studied for condition monitoring and prediction of online quality of the weld. The proposed embodiment discusses a predictive condition monitoring of a welding process, using acoustic sensors and processing their data. Converting the processed data to a predictive model, comparing different Machine Learning (ML) algorithms like Decision Tree (DT), Artificial Neural Networks (ANN) and Convolutional Neural Networks (CNN) based on the validation and training results, selecting the suitable algorithm for online or real-time weld quality prediction is the objective of the proposed embodiment. According to the research, effectively executing the proposed solution can improve its life, efficiency, and utilization. The training accuracy is observed to be 97 % for DT, 50 % for ANN and 67 % for CNN. DT, on the other hand, has a validation accuracy of 94.7 %, ANN has 47.65 % and CNN has 65.8 %. DT produced the best results in the study, initially DT prediction model is used in offline testing of the new data-set for individual weld class and the system produced 97 % accuracy. Furthermore the DT prediction model is implemented in online weld quality prediction in three classes: Bad weld, OK weld and Good weld where the system produced 90 % accuracy in real-time prediction.

Deep learning near-infrared quality prediction based on multi-level dynamic feature

As a fast, efficient, nondestructive, and pollution-free technology, near-infrared spectroscopy (NIRS) has become a practical method for online quality prediction in the process industry. The inherent dynamics of the industrial processes make the NIR spectral data no longer accord with the assumption of independent and identically distributed, which leads to the limitations of the static quality prediction models. In addition, the existing NIRS-based quality prediction methods lack consideration of the possible nonlinear relationship between NIR spectra and quality variables of complex organic samples and the over sensitivity brought by molecular-level detection. This paper proposes a multi-level dynamic feature-based deep learning NIR quality prediction method to solve the above problems. Firstly, short-term feature extraction is mainly based on 2D convolution neural networks (CNNs) to extract short-term dynamic features and eliminate the multi-collinearity and information redundancy with a 2D NIR dynamic spectral matrix input. Thereinto, considering the levels of short-term dynamic features, a combination of dilated CNN and the dense connection is proposed to construct the multi-level features and ensure the continuisty and integrity of the time dimension. Secondly, based on the extracted short-term dynamic features, the long-term dynamic spectral prediction constructed with gate recurrent unit (GRU) is proposed to capture long-term dependence between NIR spectra and predict the NIR spectrum at the next time step. Here, considering the inability of the original GRU to take account of the dynamic feature at all time steps and the possible time lag between industrial laboratory sampling and NIRS detection, temporal attention is utilized to redistribute and fuse the dynamic features of the GRU at all time steps. Finally, a fully connected layer is used to realize the regression from the NIR spectrum of the next moment to the corresponding quality variables. The effectiveness and accuracy of the method are verified by a case of the reaction and distillation process of 2,6 xylenol and corresponding NIR spectral data.

End-to-end online quality prediction for ultrasonic metal welding using sensor fusion and deep learning

In industrial-scale production applications of ultrasonic metal welding (UMW), there is a strong need for predicting joint quality quickly, reliably, and non-destructively. State-of-the-art quality assessment methods such as destructive tensile testing and binary quality classification cannot meet such requirements. This paper develops a novel end-to-end online quality prediction method for UMW based on sensor fusion and deep learning. This method first preprocesses 1-dimensional signals from multiple sensors including an acoustic emission sensor, a linear variable differential transformer, and a microphone, and transforms them to 2-dimensional images using wavelet transform. Then, these images are fed into ResNet20, which is a 20-layer convolutional neural network, to automatically generate feature maps and predict joint strength. The proposed method offers important advantages compared to state-of-the-art approaches, including automatic feature generation and good robustness to UMW tool conditions. The effectiveness of the developed method is demonstrated using real-world data generated from an UMW process with four different tool conditions. Additionally, we propose three feature fusion strategies (early fusion, middle fusion, and late fusion) and present a comparative case study to compare their performance. It is found that the late fusion strategy achieves the best prediction performance. Towards interpretability and explainability in deep learning, we perform a correlation analysis to reveal the connection between ResNet-generated features and features that are manually extracted based on UMW process physics. It is shown that many manual features are strongly correlated with ResNet features, proving that ResNet is able to resemble physical knowledge. The proposed method is readily applicable to industrial-scale UMW processes to enable accurate online quality prediction.

Domain Adaptation Mixture of Gaussian Processes for Online Soft Sensor Modeling of Multimode Processes When Sensor Degradation Occurs

Sensor degradation seriously hinders the practical application of soft sensors. To reduce the negative effect of sensor degradation, in this article, we propose a robust domain adaptation mixture of Gaussian processes (DA-MGP) for online soft sensor modeling of multimode processes. Based on the decomposition of industrial data into a group of Gaussian domains, Gaussian domain discrepancy (GDD) is designed for domain adaptation and process mode recognition. After recognizing the process mode based on GDD, a Gaussian domain adaptation is presented to correct the drifted online input data by domain mapping, which can significantly improve the robustness of the soft sensor against sensor degradation. Furthermore, the domain mapping matrix is utilized as a transferred basis function for a local transferred Gaussian process component, which is used for robust soft sensor modeling. Additionally, an online block processing framework is adopted when the DA-MGP-based soft sensor is applied in online quality prediction. Finally, the TE benchmark process and a real industrial polypropylene process are employed to verify the effectiveness of the proposed method. In the designed five cases of sensor degradation, the DA-MGP-based soft sensor shows its strong robustness against sensor degradation.

Spatial Identification of "Zeroth Defect" Formation in Organic Light-Emitting Diodes by Multispectral Mappings.

In recently developed organic semiconductors, the continuously improving sample purity makes the stability of the chemical bonds of organic materials themselves become a key factor in device stability, which provides greater uncertainty for the generation of "zeroth defect", and the spatial resolution of performance at different positions becomes particularly important. In this work, complete maps of electroluminescent, photoluminescent, and Raman scattering in the same area on an organic light-emitting diode during its operation have been achieved with a confocal spectrometer with multiple laser sources. The different spectral characters help to establish different regions and suggest the mechanism of degradation. In particular, Raman scattering has been shown to be very sensitive in a multilayer device to a change in thickness of several nanometers. In amorphous films with few defects, the very weak film uniformity, including the thickness and degree of aggregation, would induce dramatic degradation. The relatively thin and/or loosely textured region easily locally overheats and has the highest probability of "zeroth defect" generation. This method has high spatial resolution, a low level of damage to samples, good reproducibility, and multiple interconnected pieces of information, which is significant for online quality prediction and mechanistic analysis.

Two-Dimensional, Two-Layer Quality Regression Model Based Batch Process Monitoring

In this paper, a two-dimensional, two-layer quality regression model is established to monitor multi-phase, multi-mode batch processes. Firstly, aiming at the multi-phase problem and the multi-mode problem simultaneously, the relations among modes and phases are captured through the analysis between process variables and quality variables by establishing a two-dimensional, two-layer regression partial least squares (PLS) model. The two-dimensional regression traces the intra-batch and inter-batch characteristics, while the two-layer structure establishes the relationship between the target process and historical modes and phases. Consequently, online monitoring is carried out for multi-phase, multi-mode batch processes based on quality prediction. In addition, the online quality prediction and monitoring results based on the proposed method and those based on the traditional phase mean PLS method are compared to prove the effectiveness of the proposed method.

Bayesian neural network with efficient priors for online quality prediction

In process industries, neural networks are popular models for quality prediction due to their superiorities on dealing with nonlinearity and high dimensionality of process data. However, small sample size of online data will result in epistemic uncertainty of prediction model, so neural networks with fixed parameters are prone to overfitting and provide over-confident predictions with poor robustness. To this end, we propose a Pior-Assistant Random Bayesian Neural Network (PA-RBNN). First, the Random Vector Functional Link Neural Network (RVFLNN) with Bayes backpropagation (RBNN) is introduced to the last layer of the off-lined deep neural network model. The prior mean values and variances of parameters are determined by the closed-form solution from RVFLNN. Then, the semi-supervised form of PA-RBNN is further developed by introducing manifold regularization to RVFLNN to obtain the semi-supervised prior means and variances of parameters. In this way, the proposed prior provides a reasonable range before learning the parameter posterior, and both the labeled and unlabeled data are utilized to obtain the prior. The proposed method is verified by the benchmarked industrial process data. The experimental results show that the assistant prior has an optimistic effect on guiding the training process towards the basins of attraction of minima with less computation cost, and improves the generalization of prediction model.

Online quality prediction based on the double cumulative model for multiphase batch processes with transitions

In multiphase batch processes, the final product quality prediction results depend on the common cumulative effects of all the critical phases with transitions on quality rather than a separated one. However, the common cumulative effect are rarely considered at the same time. To address this issue, a double cumulative model is proposed in this paper. The double model includes the internal cumulative model and the external cumulative model. On one hand, the critical local cumulative quality is introduced to isolate the local cumulative effect of different phases and transitions in the internal cumulative model. Especially, for transitions modeling, an average process trajectories method based on the dominant phase identification is developed to deal with the dynamic and uncertainty. On the other hand, the correlation of different cumulative models is extracted by constructing the external cumulative model. In this way, the cumulative effect of all the phases and transitions are considered simultaneously. The proposed method is applied to penicillin fermentation. The simulation results demonstrate the effectiveness and superiority of the proposed method over the competing models.

Quality-Analysis-Based Process Monitoring for Multi-Phase Multi-Mode Batch Processes

In batch processing, not only the characteristics of different phases are different, but also there may be different characteristics between batches. These characteristics of different phases and batches will have different effects on the final product quality. In order to enhance the safety of batch processes, it is necessary to establish an appropriate monitoring system to monitor the production process based on quality-related information. In this work, based on multi-phase and multi-mode quality prediction, a new quality-analysis-based process-monitoring strategy is developed for batch processes. Firstly, the time-slice models are established to determine the critical-to-quality phases. Secondly, a multi-phase residual recursive model is established using each quality residual of the phase mean models. Subsequently, a new process-monitoring strategy based on quality analysis is proposed for a single mode. After that, multi-mode quality analysis is carried out to judge the relevance between the historical modes and the new mode. Further, online quality prediction is achieved applying the selected model based on multi-mode quality analysis, and an according process-monitoring strategy is developed. The simulation results show the availability of this method for multi-phase multi-mode batch processes.

Soft sensor development based on kernel dynamic time warping and a relevant vector machine for unequal-length batch processes

The unequal-length problem in batch process data directly affects the performance of data-driven soft sensors. Meanwhile, the nonlinearity and high dimensionality of batch process data make the unequal-length problem more serious, and the development of effective soft sensors for unequal-length batch processes has become a challenge. To fully address this challenge, an effective soft sensor based on kernel dynamic time warping and a relevant vector machine is proposed in this paper. The proposed soft sensor consists of trajectory synchronization and online prediction modeling. First, combining the kernel trick, we design a kernel DTW (KerDTW) algorithm to effectively solve the synchronization of unequal-length trajectories with high dimensionality and strong nonlinearity characteristics. Meanwhile, a novel synchronization performance combination index (SPCI) is proposed to realize adaptive selection of the optimal parameter of the KerDTW algorithm. Then, based on the synchronized batch trajectories using the KerDTW algorithm, an online prediction model is established using an RVM to achieve online quality prediction of nonlinear process data. The effectiveness of the proposed soft sensor is illustrated through a penicillin fermentation process.

Control and soft sensing strategies for a wastewater treatment plant using a neuro-genetic approach

During the last years, machine learning-based control and optimization systems are playing an important role in the operation of wastewater treatment plants in terms of reduced operational costs and improved effluent quality. In this paper, a machine learning-based control strategy is proposed for optimizing both the consumption and the number of regulation violations of a biological wastewater treatment plant. The methodology proposed in this study uses neural networks as a soft-sensor for on-line prediction of the effluent quality and as an identification model of the plant dynamics, all under a neuro-genetic optimum model-based control approach. The complete scheme was tested on a simulation model of the activated sludge process of a large-scale municipal wastewater treatment plant running under the GPS-X simulation frame and validated with operational gathered data, showing optimal control performance by minimizing operational costs while satisfying the effluent requirements, thus reducing the investment in expensive sensor devices.