On-line Process Monitoring Research Articles

Classification of melt pool states for defect detection in laser directed energy deposition using FixConvNeXt model

Abstract Laser directed energy deposition (L-DED) has emerged as a promising technique for rapid prototyping due to its cost-effectiveness and efficiency. However, the intricate and multi-scale physics of the process hinder its widespread application. This paper addresses the challenge by focusing on real-time identification of melt pool states to detect defects early and minimize resource wastage. To achieve this, a FixConvNeXt model was developed for fast and accurate monitoring of melt pool states. This model was trained using 5000 melt pool images captured during the printing of single-track deposits from a charge-coupled device. To evaluate its performance, FixConvNeXt was compared with other models using various metrics. Experimental results demonstrated that FixConvNeXt achieved superior performance in accurately identifying melt pool states with 99.1% accuracy, while also reducing computation burden and processing time. The mechanism of classification by FixConvNeXt was explained using gradient-weighted class activation mapping. The research findings highlight the potential application of online process monitoring in L-DED. This study lays the foundation for future development of an efficient deep learning network for automatic defect detection and feedback control.

Process monitoring method based on vine copula and transfer learning strategy

In practical industrial processes, the limited sampling time and other factors result in a scarcity of historical data for certain modes, leading to diminished generalization and accuracy of process monitoring models. To solve this problem, a process monitoring method based on vine copula-based dependence description (VCDD) and transfer learning strategy (TLVCDD) is proposed in this study. The proposed method constructs a VCDD model by utilizing target domain data and subsequently selects suitable candidate sample groups from the abundant source domain data based on this model. Candidate sample groups are transferred sequentially according to their priorities, which are quantified based on the maximum mean discrepancies between candidate sample groups and the target domain data. The VCDD model exhibiting the most superior overall performance during the transfer process is chosen and employed for online process monitoring. The effectiveness of the proposed method is demonstrated through a numerical example and Tennessee Eastman (TE) process.

Intelligent process monitoring of smart polymer composites using large area graphene coated fabric sensor.

Herein, we report the development of an online process monitoring system for vacuum-assisted resin transfer molding (VARTM) process using large area graphene coated in-situ fabric sensor. Besides imparting excellent mechanical properties to the final composites, these sensors provide critical information during the composite processing including detecting defects and evaluating processing parameters. The obtained information can be used to create a digital passport of the manufacturing phase to develop a cost-effective production technique and fabricate high-quality composites. The fabric sensor was produced using a scalable dip-coating process by coating 1-, 3- or 5-layers of thermally reduced graphene oxide (rGO) onto glass fabric surface according to the number of dips of the fabrics into GO solution. The electrical resistances from all electrode pairs were simultaneously and continuously recorded during distinct stages of the VARTM process to determine the relative conductance. During the vacuum cycle, the range of relative conductance increased with the number of coated rGO layers, with the 5-layer rGO-coated sensor showing the highest conductance range of 16.9 %. Additionally, it was observed that the 5-layer coated sensor showed a consistent decrease in conductance during the infusion phase due to the fluid flow pressure dominating the resin electrical conductivity.

Demonstrating the applicability of Raman spectroscopy for the characterization of hydrogen loading of bicyclic liquid organic hydrogen carriers (LOHCs) under process-relevant conditions

During hydrogenation and dehydrogenation reactions of liquid organic hydrogen carriers (LOHC), their thermophysical properties change significantly as a function of the degree of hydrogenation (DoH). Compared with conventional techniques, Raman spectroscopy offers a contactless methodology for online process monitoring without the need of sampling. This work demonstrates the application of Raman spectroscopy for the determination of the DoH under process-relevant conditions for LOHC systems based on diphenylmethane (DPM) and benzyltoluene (BT) up to 473 K. For the DPM-based system, a previous study is extended by examining the influence of process-like disturbances. For the BT-based system, a novel approach using indirect hard modeling of the unpolarized Raman spectra is applied, which results in strongly reduced uncertainties and insensitivity towards process-like disturbances, e.g. reaction by-products. For a corresponding testing set exhibiting a pronounced fluorescent background in the Raman spectra, the DoH is determined with an average absolute deviation of 0.0057.

Acoustic emission monitoring for industrial pressure vessels and infrastructure

Acoustic emission (AE) is a phenomenon that occurs in materials that are exposed to stress. Defects in the material, such as cracks can grow and release energy in the form of a mechanical wave that propagates through the material. With the help of piezoelectric sensors, the mechanical wave is converted into an electrical signal. The signals are recorded by the measuring system and are available for further processing. Acoustic emission as a method that records and processes the incoming signals in real time is therefore very suitable for 24/7 long-term structural health monitoring. With new approaches in data processing, acoustic emission can be moved from the classic periodic test approach to an online monitoring process. The in-service monitoring brings measuring challenges with it, like application on hot surfaces and inside explosion hazard areas. New developments in hardware and software provide tailor-made solutions for a variety of monitoring tasks. The use of artificial intelligence algorithms helps to improve the analysis of measurement data and guarantees safe operation of pressure equipment and infrastructure around-the-clock. Some examples of continuous monitoring are shown on pressure vessels from the industry sector and on infrastructure buildings e.g., steel bridges. Also, a new lean system approach in structural health monitoring with acoustic emission is presented.

Data-Driven Modeling of the Spray Drying Process. Process Monitoring and Prediction of the Particle Size in Pharmaceutical Production.

Spray drying is used in the pharmaceutical industry for particle engineering of amorphous solid dispersions (ASDs). The particle size of the spray-dried (SD) powders is one of their key attributes due to its impact on the downstream processes and the drug product's functional properties. Offline and inline laser diffraction methods can be used to estimate the product's particle size; however, the final release of these ASDs is based on offline instruments. This paper presents a novel data-driven modeling approach for predicting the particle size of SD products. The model-based characterization of the process and the product's particle size, as a critical quality attribute, follows the quality by design principles. The resulting model can be used for online process monitoring, reducing the risks of out-of-specifications products and supporting their real-time release. A Tucker3 model is trained to capture and factorize the deterministic variability of the process. Subsequently, a partial least-squares regression model is calibrated to model the impact that variability in the input material properties, the process parameters, and the spray nozzle have on the products' particle size. This strategy has been calibrated and validated using large scale production data for two intermediate drug products under high sparsity of particle size data. Despite the challenges, high accuracy was obtained in predicting the median particle size (dv50) for release. The 99% confidence interval results in an error of maximum 2.5 μm, which is less than 10% of the allowed range of variation.

Extracting powder bed features via electron optical images during electron beam powder bed fusion

Electron beam powder bed fusion offers the unique opportunity to observe the process by measuring scattered electrons on a metal detector. This technique is the state of the art in generating electron optical images of the build area after melting using single- or multi-detector setups. The images enable the detection of surface defects like porosity or material transport by reconstructing the surface topography. Internal defects such as layer-bonding defects cannot be identified. Many of these defects, particularly layer-bonding defects, often originate from an irregular distribution of the powder bed.This work introduces an additional process step by recording an electron optical image after the distribution of the powder bed. Combining this with an electron optical image after melting the previous layer enables extraction of powder bed features such as the current powder bed height. The underlying method bases on the correlation of experimental measurements and numerical simulations of the intensity of the electron optical signal for different powder bed heights. With this approach, it is possible to identify irregular powder distributions, such as uncovered areas of previously molten material or locally varying powder bed heights. This information is crucial for online monitoring and real time process control. Exemplary, this opens the opportunity of healing the powder bed by an additional raking step.

Predicting Bulk Density for Agglomerated Raspberry Ketone via Integrating Morphological and Size Metrics Using Artificial Neural Networks

The bulk density of the particles, which is directly related to transportation and storage costs, is an important basic characteristic of products as well as an important parameter in many processing systems. This work quantified the relationship between the tapped bulk density of raspberry ketone with different degrees of agglomeration and morphological metrics (particle shape descriptors and roughness descriptors) and size metrics (size descriptors) and developed an artificial neural network (ANN) prediction model for the tapped bulk density of raspberry ketone. Samples prepared under different conditions were sieved and remixed, the tapped bulk density of the particles was then measured, and the descriptor features of the particles were obtained by combining them with image processing. The dimensions of the variables were decreased by principal component analysis and variance processing. To overcome the hyperparameter estimation of the heuristic-based artificial neural networks, the network model architectures were optimized by a neural architecture search strategy combining two-objective optimization. The results demonstrated that the tapped bulk density of raspberry ketone products is not only related to the descriptors of particle size and shape but also has a non-negligible relationship with particle roughness descriptors. The performance of the optimal ANN model demonstrated that the model can well predict the tapped bulk density of raspberry ketone with different degrees of agglomeration. The ANN model obtained by extracting morphology and size metrics through online image analysis can be used to measure the tapped bulk density in real-time and has the potential to be used for developing model-based online process monitoring.

Knowledge distillation-based information sharing for online process monitoring in decentralized manufacturing system

Knowledge distillation-based information sharing for online process monitoring in decentralized manufacturing system

Monitoring epidemic processes under political measures.

Statistical modeling of epidemiological curves to capture the course of epidemic processes and to implement a signaling system for detecting significant changes in the process is a challenging task, especially when the process is affected by political measures. As previous monitoring approaches are subject to various problems, we develop a practical and flexible tool that is well suited for monitoring epidemic processes under political measures. This tool enables monitoring across different epochs using a single statistical model that constantly adapts to the underlying process, and therefore allows both retrospective and on-line monitoring of epidemic processes. It is able to detect essential shifts and to identify anomaly conditions in the epidemic process, and it provides decision-makers a reliable method for rapidly learning from trends in the epidemiological curves. Moreover, it is a tool to evaluate the effectivity of political measures and to detect the transition from pandemic to endemic. This research is based on a comprehensive COVID-19 study on infection rates under political measures in line with the reporting of the Robert Koch Institute covering the entire period of the pandemic in Germany.

Medium Optimization and Fermentation Kinetics for Antifungal Compounds Production by an Endophytic Paenibacillus polymyxa DS-R5 Isolated from Salvia miltiorrhiza.

An endophytic bacterium Paenibacillus polymyxa DS-R5 which can effectively inhibit the growth of pathogenic fungi was isolated from Salvia miltiorrhiza in our previous study. By using hydrochloric acid precipitation, methanol extraction, silica gel column isolation, dextran gel chromatography column, and HPLC, 3 compounds with antifungal activity were isolated. To further improve the production of antifungal compounds produced by this strain, fermentation medium was optimized using one-factor-at-a-time, Plackett-Burman design, and Box-Behnken design experiments. Through statistical optimization, the optimal medium composition was determined to be as follows: 14.7g/l sucrose, 20.0g/l soluble starch, 7.0g/l corn steep liquor, 10.0g/l(NH4)2SO4, and 0.7g/lKH2PO4. In this optimized medium, the highest titer of antifungal compounds reached 3452 U/ml, which was 123% higher than that in the initial medium. In addition, in order to guide scale-up for production, logistic and Luedeking-Piret equations were proposed to predict the cell growth and antifungal compounds production. The fermentation kinetics and empirical equations of the coefficients (X0, Xm, μm, α, and β) for the two models were reported, which will aid the design and optimization of industrial processes. The degrees of fit between calculated values of the model and the experimental data were 0.989 and 0.973, respectively. The results show that the cell growth and product synthesis models established in this study may better reflect the dynamic process of antifungal compounds production and provide a theoretical basis for further optimization and on-line monitoring of the fermentation process.

Deep learning-based image analysis for in situ microscopic imaging of cell culture process

Mammalian cell culture is an important bioprocess that directly affects the quality and yield of biopharmaceuticals. Traditionally, condition monitoring of the operation is based on sampling periodically and off-line analysis, which is labor intensive, time consuming, and causing time delays. In this work, in situ microscope is investigated for on-line real-time monitoring of the culture process of Chinese hamster ovary cells with focus on investigation of deep learning-based Mask R-CNN algorithm for image analysis. The model is trained by 184 images with 183,040 cells using data augmentation methods and transfer learning technique. Mask R-CNN segmented the clustered cells more effectively than the conventional one combining edge detection, intensity thresholding, and advanced watershed method as well as the multi-scale edge detection method. Its Dice score, accuracy, precision, sensitivity, F1 score, specificity, and relative volume difference reach 0.862, 0.945, 0.901, 0.827, 0.862, 0.977, and 0.082, respectively. The evolution of geometrical features of cells were further analyzed, including equivalent diameter, circularity, aspect ratio, and eccentricity. The result demonstrated the great potential of deep learning technology in analysis of on-line images for optimization and control of the cell culture process.

Streaming variational probabilistic principal component analysis for monitoring of nonstationary process

Modern industrial processes are characteristic of nonstationary and uncertainty. To address these issues, this paper proposes a probabilistic principal component analysis based model that utilizes streaming variational inference for online monitoring of nonstationary process. The model parameters are updated in a streaming/batch-wise manner to track the nonstationary behavior in the process. In the updating procedures, the posterior distributions obtained from the previous data batch are used as priors for the current model, whose parameters can be estimated using the variational Bayesian inference. The streaming variational inference enables recursive parameter estimation over time. Additionally, in order to improve computational efficiency, a distributed update strategy is also applied to the proposed model. For online process monitoring, the noncentral chi-square distributions are derived to estimate the statistical control limits for each data batch, resulting in better adaptation to the nonstationary characteristics. The effectiveness and superiority of the proposed method are validated through simulation example and real-world industrial application.

Explainable Remaining Tool Life Prediction for Individualized Production Using Automated Machine Learning.

The increasing demand for customized products is a core driver of novel automation concepts in Industry 4.0. For the case of machining complex free-form workpieces, e.g., in die making and mold making, individualized manufacturing is already the industrial practice. The varying process conditions and demanding machining processes lead to a high relevance of machining domain experts and a low degree of manufacturing flow automation. In order to increase the degree of automation, online process monitoring and the prediction of the quality-related remaining cutting tool life is indispensable. However, the varying process conditions complicate this as the correlation between the sensor signals and tool condition is not directly apparent. Furthermore, machine learning (ML) knowledge is limited on the shop floor, preventing a manual adaption of the models to changing conditions. Therefore, this paper introduces a new method for remaining tool life prediction in individualized production using automated machine learning (AutoML). The method enables the incorporation of machining expert knowledge via the model inputs and outputs. It automatically creates end-to-end ML pipelines based on optimized ensembles of regression and forecasting models. An explainability algorithm visualizes the relevance of the model inputs for the decision making. The method is analyzed and compared to a manual state-of-the-art approach for series production in a comprehensive evaluation using a new milling dataset. The dataset represents gradual tool wear under changing workpieces and process parameters. Our AutoML method outperforms the state-of-the-art approach and the evaluation indicates that a transfer of methods designed for series production to variable process conditions is not easily possible. Overall, the new method optimizes individualized production economically and in terms of resources. Machining experts with limited ML knowledge can leverage their domain knowledge to develop, validate and adapt tool life models.

Online cloud management and quality monitoring of the whole process of undergraduate thesis

Abstract The undergraduate thesis is an important practical teaching part of the undergraduate teaching program, and it is important to manage its quality. In this paper, an online cloud management system for the thesis is designed according to the requirement analysis, and the ARIMA model in time series is used to predict the number of thesis supervisors by the supervisor and give a reminder for the supervisor’s guidance decision. For the risk management of dissertation progress, a Bayesian network-based assessment method is proposed. From the perspective of dissertation similarity, a process of dissertation quality analysis based on the SimHash algorithm is constructed. The average error rate of ARIMA is only 0.190, which is optimized to a certain extent compared with other algorithms. 20 samples of the inference operation results have no sample with the probability of the dissertation progress delay level at High below 40%.

Quality Evaluation and Analysis of Force Signals in CFRP Helical Milling

Nowadays, drilling is one of the most widespread operations due to the need to obtain holes for mechanical joints. This operation is particularly relevant in the aeronautical sector due to the high number of operations performed, the high demands and the use of materials that are difficult to machine. The use of materials that enhance aircraft performance, such as carbon fiber composites (CFRP), poses a challenge for axial drilling operations. The characteristics of these materials and typical defects such as delamination are difficult to control due to the axial forces produced during drilling operations. Therefore, more efficient alternatives are required. In this context, helical milling operations have advantages such as lower axial stress, higher flexibility, and higher efficiency in heat and chip evacuation that put it in the spotlight. In this work, research has been carried out in which helical milling operations have been performed on CFRP. The main objectives were the analysis of surface quality, delamination, tool wear and force analysis. For this purpose, a working methodology has been developed combining machining parameters that define the kinematics of the cutting process (cutting speed, axial feed rate, and tangential feed rate). Finally, this information has allowed finding a correlation between quality indicators such as delamination and the forces generated during the cutting process and associated with progressive tool wear. The results show that the force signal could be used for on-line monitoring of the machining process.

Alarm-based explanations of process monitoring results from deep neural networks

Deep Learning (DL) models are becoming the preferred approach for process monitoring due to their higher prediction accuracy; however, they are still viewed as black boxes. Explainable Artificial Intelligence (XAI) methods seek to address this shortcoming by providing explanations that are either global (explains the entire DL model) or local (explains the result for each individual sample). Due to nonlinearities and other complexities inherent in chemical processes, providing a local explanation is more suitable. This paper proposes a local XAI method that explains process monitoring results from a deep neural network (DNN) based on process alarms. The effectiveness of the proposed method is demonstrated using a CSTR case study and the Tennessee-Eastman benchmark process. Our results show that the explanations proffered by the proposed method assist operators in understanding DNN's prediction during online process monitoring. Additionally, during model development, it can offer insights that enable improvement of the DNN's architecture.

Online monitoring of direct laser metal deposition process by means of infrared thermography

Direct laser metal deposition (LMD–DED) is an additive manufacturing (AM) process that is used to build up and repair high-quality metal components. It works by overlapping layers of powder material and melting them with a laser. To get a stable process without defects and to reach, at the same time, high mechanical properties, a robust assessment and control of the process parameters, and above all of their combination, is required. The ideal goal is to assure the online control, to stop or correct the process in case of unexpected anomalies. In this work, a robust online monitoring of the laser metal deposition (LMD–DED) process based on the use of infrared thermography was developed and proposed. After choosing the suitable process parameters, a customized design of experiments (DOE) was set, and the statistical analysis of different thermal features was carried out to develop the most robust models that correlate them with the input process parameters (laser power, scanning speed, and powder flow rate). The proposed procedure was based on the extraction of different thermal features from suited regions of interest (ROI), performing statistical analyses by means of analysis of variance (ANOVA) and building regression models to correlate the process parameters with the thermal behavior. The obtained results demonstrated the possibility to control the process by means of the chosen thermal features, independent of the position of the ROI. Moreover, the possibility to use the models to detect typical AM defects, and anomalies, online directly during the process, has been proved and verified by destructive macrographs carried out on the manufactured coupons.

On-line monitoring of a coal separation process in a jig – a simulation study

On-line monitoring of a coal separation process in a jig – a simulation study

Online process monitoring of direct laser interference patterning using an infrared camera system

Direct laser interference patterning enables the fabrication of homogeneous microstructures in the sub µm range. Generally, the topography of fabricated structures is characterized ex-situ using for instance confocal microscopy or white light interferometry. These methods require at least few seconds to fulfill a measurement and therefore are not suitable for online process observation. In this study, an IR camera system is used for the first time, for capturing the average temperature during the DLIP treatment in real time. To determine a correlation between the average temperature and the topographical parameters, the treated surfaces are analyzed using confocal microscopy. A linear dependency between the applied laser fluence (in the range from 0.7 to 4.9 J/cm2) and the average measured temperature could be established. In addition, in this fluence range, significant changes in the surface roughness, skewness and kurtosis occurred, which could be associated to a certain process average temperature.