The goal of this paper is to make a strong point for the usage of dynamical models when using reinforcement learning (RL) for feedback control of dynamical systems governed by partial differential equations (PDEs). To...
详细信息
ISBN:
(纸本)9798331540920;9783907144107
The goal of this paper is to make a strong point for the usage of dynamical models when using reinforcement learning (RL) for feedback control of dynamical systems governed by partial differential equations (PDEs). To breach the gap between the immense promises we see in RL and the applicability in complex engineering systems, the main challenges are the massive requirements in terms of the training data, as well as the lack of performance guarantees. We present a solution for the first issue using a data-driven surrogate model in the form of a convolutional Long-Short Term Memory network with actuation. We demonstrate that learning an actuated model in parallel to training the RL agent significantly reduces the total amount of required data sampled from the real system. Furthermore, we show that iteratively updating the model is of major importance to avoid biases in the RL training. Detailed ablation studies reveal the most important ingredients of the modelingprocess. We use the chaotic Kuramoto-Sivashinsky equation do demonstrate our findings.
processcontrol has been established as a core course for the formation of chemical engineers. Very often, it is the only course dealing with the analysis of transient (time dependent) phenomena and conditions. It rel...
详细信息
processcontrol has been established as a core course for the formation of chemical engineers. Very often, it is the only course dealing with the analysis of transient (time dependent) phenomena and conditions. It relies on difficult concepts requiring intensive mathematical approaches and simulations based on differential equations and Laplace transform. It is commonly criticized for its level of abstraction and mathematical involvement, in contrast to other courses in the career, and for the restricted applicability to industrial jobs. This criticism generally negatively affects the motivation of students. However, the combination with hands-on experiments has proved to enrich the learning and motivation of students, but most colleges face severe restrictions on the investment, maintenance, and operation of processcontrol labs and the addition of new requirements in the curriculum. Some alternatives have been exploring the use of simple modules for classroom demonstrations, theoretical simulations of equipment in unit operations lab, and virtual-lab simulations. This paper describes the scope of technical training based on process model and synthesis of PID controllers for six experimental set-ups with liquid level and temperature control, using lab equipment fully automated for data acquisition, handling of manipulated and disturbance variables, and selection of parameters for PID controllers. MATLAB codes and Simulink graphical simulations support the processing of data and analysis of results. In addition, the course develops a unique experience in team skills and performance where every team is a combination of two sub-teams. The "office" sub-team oversees research on industrial applications, instrumentation characteristics, and computational modeling. The "lab" sub-team oversees elaborating and testing experimental plans, collecting data, and analyzing results. Every team is assigned two sequential projects;one for processmodeling (open-loop) and one for co
In this article we introduce STARDUST (event STream analysis for process Discovery Using Sampling sTragies), a process discovery approach that analyses a trace stream, in order to discover a process model that may cha...
详细信息
In this article we introduce STARDUST (event STream analysis for process Discovery Using Sampling sTragies), a process discovery approach that analyses a trace stream, in order to discover a process model that may change over time. The basic idea is to adopt a sampling technique to select the most representative trace variants to be considered for the process discovery, then to alert a concept drift as the trace variants to be sampled change over time and, finally, to trigger the discovery of a new process model as a drift is alerted. We formulate the proposed approach under the assumption that the trace distribution commonly follows the Pareto's principle (i.e., a few trace variants covers the majority of cases) which is commonly satisfied in several business processes. Experimental results on various benchmark event logs handled as streams show the effectiveness of the proposed approach also compared to a state-of -the-art concept drift detection approach.
With the rapid increase in the complexity of electronics, the cost of the functional testing process used to ensure product functionality continues to rise. Optimization modeling based on reliability analysis is an ef...
详细信息
control of nonlinear dynamical systems is a complex and multifaceted process. Essential elements of many engineering systems include high-fidelity physics-based modeling, offline trajectory planning, feedback control ...
详细信息
control of nonlinear dynamical systems is a complex and multifaceted process. Essential elements of many engineering systems include high-fidelity physics-based modeling, offline trajectory planning, feedback control design, and data acquisition strategies to reduce uncertainties. This article proposes an optimization-centric perspective which couples these elements in a cohesive framework. We introduce a novel use of hyper-differential sensitivity analysis to understand the sensitivity of feedback controllers to parametric uncertainty in physics-based models used for trajectory planning. These sensitivities provide a foundation to define an optimal experimental design which seeks to acquire data most relevant in reducing demand on the feedback controller. Our proposed framework is illustrated on the Zermelo navigation problem and a hypersonic trajectory control problem using data from NASA's X-43 hypersonic flight tests.
The article discusses an approach aimed at the formation of a methodology for applying the principles and models used in the automatic control theory in relation to the processes of enrichment of mineral raw materials...
详细信息
The article discusses an approach aimed at the formation of a methodology for applying the principles and models used in the automatic control theory in relation to the processes of enrichment of mineral raw materials. The problem solved in the article is posed as follows, try to represent the process of gravitational enrichment, namely centrifugal concentration, in the form of a transfer function and try to find its parameters. The paper proposes an approach to modeling centrifugal concentrators based on transfer functions, using averaged values of output characteristics. Proposal of a mechanism for determining the parameters of transfer functions and their calculation based on experimental data. To check the presented calculations, the MatLAB software tool with the Simulink package was used. As a result, the transfer function of the KC-CVD6 apparatus used in the gold deposit was found and satisfactory results were obtained. The work uses methods of identification of automatic systems, methods of system analysis, as well as methods of mathematical modeling, namely the numerical solution of differential equations, implemented in the MatLAB environment with the Simulink package. A comparison with the classical regression model was made;as a result, the modeling error was reduced by two times. The simulation error for the concentrate yield was less than 1%.
The cement specific surface area is an important indicator of cement quality. The accurate prediction of the cement specific surface area aims to guide operators to control the cement grinding process to improve produ...
详细信息
The cement specific surface area is an important indicator of cement quality. The accurate prediction of the cement specific surface area aims to guide operators to control the cement grinding process to improve product quality while reducing system energy consumption. However, due to the complexity of the cement grinding process, the process variables have coupling, time-varying delay, nonlinear characteristics, and different sampling frequency. Herein, we proposed the specific surface area prediction model, which combined dual-frequency principal component analysis and extreme gradient boosting (DF-PCA-XGB). In order to solve the problem of difficulty in modeling due to different sampling intervals of related data, this paper analyzes the low-frequency sampling data and high-frequency sampling data under multiple working conditions, and establishes prediction models respectively. Aiming at the data redundancy problem of high-frequency and low-frequency variable data in the introduced time window, a method based on the combination of principal component analysis (PCA) and extreme gradient boosting (XGB) cross-validation is proposed to reduce data redundancy while retaining most of the characteristics of the data. The final specific surface area prediction results were obtained by weighting the high-frequency data model and the low-frequency data model. The simulation results showed that the prediction method in this paper can improve the prediction accuracy of the specific surface area of the finished cement product under multiple working conditions with high stability and has promising application in the cement manufacturing process.
The shifted (or two-parameter) exponential distribution is a probability model that is widely used in many practical applications to model time-to-event data. In reliability analysis it has been frequently used for mo...
详细信息
The shifted (or two-parameter) exponential distribution is a probability model that is widely used in many practical applications to model time-to-event data. In reliability analysis it has been frequently used for modeling the lifetime of products with a warranty period. In this paper, we propose and study the properties of a run sum chart for monitoring shifted exponential lifetimes in order to detect a change in either or in both process parameters. Using the Markov chain method, we evaluate several performance measures, based on the run length distribution of the proposed chart, and investigate its performance for increasing and/or decreasing shifts in process parameters. The results of an extensive numerical study show that a properly designed run sum chart has an improved detection ability compared to that of a Shewhart-type Max chart for shifted exponential distribution. In addition, the proposed chart can be considered as a viable alternative to the CUSUM-type Max chart for shifted exponential distribution, since for a large number of out-of-control situations it requires approximately the same, if not less, time to detect them. Finally, for supporting the use of the proposed run sum chart in practice, we provide guidelines and empirical rules for selecting the values of its parameters, along with an illustrative numerical example.
Advanced manufacturing processes are often based on complex multiphysics phenomena that are either poorly understood or are computationally too expensive to simulate in the context of process design, control, or plann...
详细信息
Advanced manufacturing processes are often based on complex multiphysics phenomena that are either poorly understood or are computationally too expensive to simulate in the context of process design, control, or planning. Traditionally, simplified physics models with prescribed heuristics or purely data-driven surrogate models are used as alternatives in such applications. The concept of physics-informed machine learning (PIML) has been shown to have unique advantages over both of these alternatives in various fields of complex system analysis. In this paper, a new PIML approach is presented to model the geometry of the cut produced by a magnetically assisted laser-induced plasma micro-machining (M-LIPMM) process. This PIML architecture uses a neural network to auto-adapt the parametric boundary condition and physical properties used in a simplified finite difference-based physics model (of 2D heat conduction), as a function of the inputs namely the laser settings. This network also estimates the scaling and shifting parameters used by a convolutional neural network that takes the temperature profile predicted by the simplified heat conduction model to predict the width and depth of the machined cut. Trained on physical experiment data, the PIML approach compares favorably to a pure data-driven neural network model in extrapolation tests, while also providing physical insights (that the latter cannot). The PIML approach also provides an 85% better accuracy overall compared to the simplified physics model with heuristic settings.
Adaptive graph neural networks (AGNNs) have achieved remarkable success in industrial process soft sensing by incorporating explicit features that delineate the relationships between process variables. This article in...
详细信息
Adaptive graph neural networks (AGNNs) have achieved remarkable success in industrial process soft sensing by incorporating explicit features that delineate the relationships between process variables. This article introduces a novel GNN framework, termed entropy-regularized ensemble adaptive graph (E(2)AG), aimed at enhancing the predictive accuracy of AGNNs. Specifically, this work pioneers a novel AGNN learning approach based on mirror descent, which is central to ensuring the efficiency of the training procedure and consequently guarantees that the learned graph naturally adheres to the row-normalization requirement intrinsic to the message-passing of GNNs. Subsequently, motivated by multi-head self-attention mechanism, the training of ensembled AGNNs is rigorously examined within this framework, incorporating an entropy regularization term in the learning objective to ensure the diversity of the learned graph. After that, the architecture and training algorithm of the model are then concisely summarized. Finally, to ascertain the efficacy of the proposed E(2)AG model, extensive experiments are conducted on real-world industrial datasets. The evaluation focuses on prediction accuracy, model efficacy, and sensitivity analysis, demonstrating the superiority of E(2)AG in industrial soft sensing applications.
暂无评论