Keywords: Data Mining & Data Analytics, Signal Processing
Abstract: Baseline shift, caused by interference factors, seriously contaminates the essential information of the further analysis. The current baseline correction methods maintain a constant smoothing parameter during the iterative calculation, while the real signal is composed of multiple intervals with different smoothness. To address this problem, a local smoothing optimization penalized least squares (lsoPLS) method is presented in this work. The local smoothing optimization strategy using the derivative difference between the previously fitted baseline and the original signal is used to identify the interval location, and smoothing parameter is optimized by the interval characteristics. Two simulated cases and one real experimental data confirm that the proposed method has good performance in the multi-interval baseline correction and can be served as an effective pre-processing method for the complex spectral signal.

Keywords: Fault Detection & Diagnosis, Data Mining & Data Analytics, Process Control & Automation
Abstract: As a measurement for the production performance, the online quality variables from soft sensors contribute greatly to obtaining immediate information from the process. The complex correlations between large numbers of process variables and disturbances inherited from the dynamic and nonlinear characteristics of chemical processes put more challenges in constructing the soft-sensor models. The soft sensors which are typically developed in steady-state conditions are not suitable for doing predictions in a dynamic operating system. This paper proposes a semi-supervised latent dynamic variational autoencoder (S2-LDVAE) to learn features between the process and quality data. Furthermore, the issue of the uneven length of the process and quality data is noteworthy. When there are fewer quality data than the process data, severe degradation to the performance of the trained model may occur. The process and quality data are encoded into the latent space in a temporal way for dynamic feature extraction. In the case of missing quality data, the artificial data generated by the trained prediction network are used to provide quality estimates during on-line prediction. The proposed method is compared with the other methods to show its contribution and performance in terms of quality prediction in a numerical case and an industrial case.

Keywords: Process Control & Automation, Intelligent Control, Identification & Estimation
Abstract: This study provides a comprehensive approach to effectively deploy end-to-end soft sensors in live industrial settings. The approach provides process-industry specific steps for data pre-processing, algorithm selection, and deployment. This study is concluded with three industrial examples of soft sensors deployed for three different applications. Each soft sensor was deployed either directly into the distributed control system (DCS) or as an application running on a dedicated local area network workstation and communicated to the DCS via OPC, for closed-loop control.

Keywords: Fault Detection & Diagnosis, Identification & Estimation, Intelligent Control
Abstract: This paper develops an extended Kalman filter (EKF) for both normal and fully oxygen-starved polymer electrolyte membrane fuel cells (PEMFCs). The underlying model for the EKF is a lumped version of a pseudo-2D model of a fuel cell. This paper serves as a proof of concept for using an EKF to accurately predict the output voltage of the fuel cell based on the input current density, impurities in the anode, and oxygen starvation. The results show a high agreement between the pseudo-2D model and the EKF model for the normal operation of a fuel cell. However, this agreement is weaker in the case of a fully starved fuel cell operation.

Keywords: Data Mining & Data Analytics, Process Control & Automation, Air Pollution Modeling & Control
Abstract: A video-based flare image monitoring system is developed for real-time estimation of the flare gas flow rate at the edge. Depending on the desired trade-off between speed and accuracy, either an object detection (EfficientDet Dx) or instance segmentation (Mask R-CNN) model is used for real-time detection of flare and smoke instances in the input video stream. Historical and synthetic data is used to achieve high precision and recall (greater than 0.98) for both flare and smoke. The detected rectangular bounding boxes or polygon masks are used to estimate the flame size, and predict the flare gas exit velocity or equivalently flow rate. The estimated flow rate is within +/- 10% of a reference flow meter. The Deep Learning models are “edgified” in order to shrink the size and improve the inference speed by ~3X on small footprint edge devices. The deep learning models of the flame size are combined with first principles knowledge to estimate the flare volumetric flow.

Keywords: Data Mining & Data Analytics, Identification & Estimation, Process Control & Automation
Abstract: This manuscript describes an inferential to estimate fuel oil viscosity on real-time basis using a hybrid modeling approach. Various conventional and machine learning approaches were compared to select the best model. The deployed inferential can be used as an advisory tool (open-loop) for refinery personnel to capture sub-optimal components distribution across the fuel oil network, or it can be implemented as part of the blending system for closed-loop control applications. A modern inferential framework was established which allows deployment of conventional or modern machine learning models in the plant control network.

Keywords: Adaptive & Learning Systems, Data-Driven Control, Fault Detection & Diagnosis
Abstract: Devices used in additive manufacturing in fast prototyping frequently have faults and issues that device operators are unaware of. Failures that go undetected may damage the device and cause the manufactured product to be faulty, thus costing additional time and resources. The research objective is to design a web-based application that monitors 3D printing operations and predicts future printer temperature values. This paper compares and contrasts various deep learning (DL) algorithms, including the multilayer perceptron (MLP), long short term memory (LSTM), and convolutional neural network (CNN). There will also be a comparative analysis of the various DL algorithms for predicting future temperature values. The system will use a web-based application connected to an Internet of Things (IoT)-based system in charge of data collection from multiple sensors attached to the device. When the model is tested, forecasting measures such as the root mean square error (RMSE), mean average error (MAE), mean absolute percentage error (sMAPE), and r-square (R^2) metrics will be examined. Based on the results of the various experiments conducted, the use of LSTM is observed to perform the best, with an RSME of 0.4062, MAE of 0.2176, sMAPE of 0.1037, and R^2 of 0.7095.

Keywords: Process Integration & Optimization, Data Mining & Data Analytics
Abstract: A distributionally robust chance-constrained programming (DRCCP) approach based on the deep kernel ambiguity set is proposed in this paper. The kernel ambiguity set possesses notable advantages over other existing ambiguity sets from the literature, and it is constructed by using the kernel mean embedding (KME) and the maximum mean discrepancy (MMD) between distributions. In the proposed method, the worst-case Conditional Value-at-Risk (CVaR) approximation is employed to approximate the distributionally robust joint chance constraint (DRJCC). Additionally, the performance of the presented method can be significantly enhanced by using the multi-layer deep arc-cosine kernel (MLACK), compared to the use of shallow kernels. The presented DRCCP approach is applied to a numeral example and a nonlinear process optimization problem to demonstrate its efficacy.

Keywords: Adaptive & Learning Systems, Data-Driven Control
Abstract: For nonlinear continuous-time systems, a robust adaptive output feedback control method based on output feedback exponential passivity (OFEP) with the cerebellar model articulation controller (CMAC) based parallel feedforward compensator (PFC) and feedforward (FF) input has been proposed. In this method, by applying the CMAC, which has low calculation cost and has strong nonlinearity characteristics, to PFC design and feedforward input design, stable output regulation control system is obtained for uncertain nonlinear systems. In this paper, we try to verify the effectiveness of the adaptive output feedback method with adaptive PFC and FF input via CMAC by experiments through a magnetic levitation system.

Keywords: Data-Driven Control, Robotics & Mechatronics, Adaptive & Learning Systems
Abstract: Model-based development (MBD) that utilizes system models in the design processes of complex products such as autonomous driving vehicles has been receiving increased attention. Moreover, an advanced control system design scheme is required to accurately control the developed products under any hard conditions in practical usage. This study proposes a control system for integrating a data-driven controller (DDC) with a model-based control (MBC) system design. The proposed method considers a hierarchical control structure consisting of an upstream control system based on the MBC design approach and a downstream control system including a plant control loop with a DDC. In this paper, the design scheme of unifying models and data in the MBD process is called smart MBD (S-MBD); in other words, the proposed control system design scheme is based on the S-MBD approach. The effectiveness of the proposed hierarchical control system was verified by applying it to a hydraulic excavator.

Keywords: Data-Driven Control
Abstract: The short period of the vehicle development is required. For this purpose, a control method for dynamometers has been proposed to perform load tests that simulate driving loads. Specifically, in the conventional method, control parameters are calculated based on system identification, and high system identification accuracy is required. In this paper, a data-driven control method that calculates the control parameters directly without system identification is proposed. In the proposed scheme, the least-squares method can be applied, and the control parameters can be obtained considering the robustness.

Keywords: Data Mining & Data Analytics, Process Integration & Optimization, Process Control & Automation
Abstract: This paper focuses on the performance assessment of various hyperparameter tuning techniques and algorithms used by Long Short-Term Memory (LSTM) networks in fore- casting uncertain parameters. An energy intensive industry installed with a hybrid energy system consisting of solar photo voltaic (PV) panels, waste to energy (WTE) plants and main electricity grid is considered as a case study for demonstration. Among the three case studies considered for hyperparameter tuning, the first case study deals with a manual tuning approach. While the second one focuses on a tuning strategy using Optuna with grid search algorithm, the third case study concentrates on adopting Optuna with a Bayesian optimization framework. The overall objective entails generating an optimal forecast of the uncertain parameters (in this case solar irradiance and main grid electricity price) using LSTM adopting various hyperparameter tuning techniques in order to solve a day-ahead energy scheduling Mixed Integer Linear Programming (MILP) cost and carbon emission minimization problem.

Keywords: Robotics & Mechatronics, Intelligent Control, Adaptive & Learning Systems
Abstract: The main goal of the teleoperation systems is to achieve the highest transparency possible while maintaining stability in the presence of the time delay. This paper proposes the adaptive energy reference Time Domain Passivity Approach (TDPA) to teleoperation systems with communication delays, in order to overcome the drawbacks of the conventional TDPA such as sudden force change, conservatism, and position drift. The proposed method establishes a reference energy function by estimating the passive elements of the system, i.e., eliminating the active parts. The passive elements are estimated by utilizing the Recursive Least Square (RLS) method. The controller makes the system follow the reference energy by dissipating energy with a variable damping element. Since the controller is activated once the energy decreases, it has smoother force changes than the conventional TDPA which is activated once energy gets negative. The adaptive energy reference TDPA is extended for teleoperation systems in a structure that can avoid the position drift and the conservative passivity condition of the conventional TDPA by eliminating the parallel passivity controller. The simulation results show that the adaptive energy reference TDPA reduces the force changes up to 23~percent in comparison with the conventional TDPA. In addition, the energy of the adaptive energy reference TDPA is less dissipated than the conventional TDPA which shows a less conservative behavior. Furthermore, no position drift is observed in the adaptive energy reference TDPA.

Keywords: Control of Cyber-Physical Systems, Data Mining & Data Analytics, Identification & Estimation
Abstract: This paper proposes a modified consensus-based distributed algorithm to solve a class of convex network optimization problems. In particular, we consider the directed multi-agent networks, where the communications are not necessarily bidirectional. Under standard convexity and smoothness assumptions about the local cost functions and strong connectivity of the communication network, we show that the proposed network flow converges exponentially to the global minimizer. Simulation studies are also provided to corroborate our results.

Keywords: Identification & Estimation, Data Mining & Data Analytics, Signal Processing
Abstract: Slow-feature analysis (SFA) seeks to extract the most slowly varying components of dynamic systems. However, the original definition of SFA implies a linear relationship of system states between adjacent time instants. In this paper, a new approach to SFA, which is called EVOLVE·INFOMAX, is defined, based on which the mutual-information-based SFA (MI-based SFA) is proposed. The optimisation problem can be solved by joint diagonalisation of the mutual-information (MI) matrices. The MI matrices are approximated by quantities related to Rényi entropy that can be calculated using the kernel trick. The case studies show MI-based SFA is better for slow feature extraction, especially for nonlinear systems. This allows a better soft sensor to be developed.

Keywords: Identification & Estimation, Signal Processing, Process Control & Automation
Abstract: The set-membership estimation problem is studied for a class of industrial processes, which is modelled by discrete linear parameter-varying (LPV) systems with uncertain scheduling parameters. Considering the unknown-but-bounded (UBB) process and measurement noise, a sufficient condition for the existence of the set-membership estimation is given with the norm-bounded uncertainty of the scheduling parameters. A convex optimization method for calculating the ellipsoidal feasible state set with guaranteed true states is proposed by solving a time-varying linear matrix inequality (LMI). Finally, a CSTR simulation shows the effectiveness of the proposed approach.

Keywords: Safe Process Operating Systems, Data Mining & Data Analytics, Data-Driven Control
Abstract: Alarm floods pose a serious threat to the safety of complex industrial plants by overloading an operator's cognitive abilities with a large number of alarms in a short period of time. Therefore, the development of methods to assist operators in handling alarm floods is of great importance. In this paper, an operator assistance system is developed that relies on similarity analysis of alarm floods and alarm scoring. A vector representation called the Modified Bag of Words is proposed to turn alarm floods into feature vectors, which are then used in a clustering algorithm for similarity analysis. An alarm weighting strategy reflecting the key features of alarm floods, such as temporal information, is proposed, which provides alarm ranking to assist operators in identifying alarms relevant to specific abnormal situations. Utilizing the Tennessee Eastman process benchmark, a qualitative assessment of the proposed approach is conducted.

Keywords: Process Control & Automation, Controller Performance Evaluation
Abstract: A method to design fractional order controllers for fractional order processes using the direct synthesis (DS) approach is proposed. The DS method is one of the most commonly used technique for tuning proportional-integral-derivative (PID) controllers. However, this approach has not been used for fractional order models. In this work, the DS method is adopted to take advantage of its simplicity; moreover, the procedure results in the structure of the controller as well as the controller parameters and the fractional orders of controller components. The resulting fractional order controllers are similar to the tilt-integral-derivative (TID) controller. Mathematical formulations are provided to obtain controllers for models with different degrees of the denominator polynomial; time-delay models are also considered. Simulation results are presented to demonstrate the efficacy of the designed controllers.

Keywords: Data Mining & Data Analytics, Fault Detection & Diagnosis, Process Control & Automation
Abstract: Modern data acquisition systems allow collection and storage of huge data from process plants. The data set includes sensor readings for the process variables, alarms and events (A&E) log which contains historical alarm data for the plant as well as non-numerical linguistic records of operator interventions and actions. Process data, A&E log, operational records and piping and instrumentation diagram (P&ID) contain wealth of information about the process. Often this information remain under-utilized due to the unavailability of appropriate data mining tools. This paper proposes a comprehensive cloud-based framework that facilitates data collection from various sources into an integrated database, process and analyze data, and display the results in precise visual forms. The application has the capability to analyze huge sets of data in a computationally efficient way. There are four main functions, which are ``Data Selection'' to select desired processing unit, variables, time range, and parameters tuning; ``Visualization'' displays the results of the analysis done on process measurements and A&E log data for specified time period; ``Alarm Design'' allows the user to re-design an alarm configuration for a specific process variable; ``Causality Analysis'' discovers the relationships between process variables and shows the root cause of an event involving multiple alarms. An industrial case study has been utilized to illustrate the effectiveness and practicability of the platform.

Keywords: Control in Medicine & Health Monitoring, Controller Performance Evaluation
Abstract: The COVID-19 pandemic is characterized by system instability, nonlinear dynamics, significant delays and large uncertainty. This combination makes it challenging to design robust feedback controlled mitigation strategies. We proposed a linearizing variable transformation that translates the challenging nonlinear problem to linear control of an integral system with time delay. In the transformed problem, robust linear controller design and analysis tools can used, and this system can be made robust to significant uncertainty. So what is the catch? This paper compares the achievable performance to a gainscheduling approach. We did not find significant limitations to performance as a result of the transformation. Locally, the same performance can be achieved. In transients between equilibrium points, the variable transformation approach maintains robust stability and performance guarantees.

Keywords: System Behaviours, Identification & Estimation, Control in Medicine & Health Monitoring
Abstract: Increasing E.Coli – AMR cases in hospitals have been a major healthcare concern in many countries, especially United Kingdom (UK). Although the current policies in the UK are helpful in early screening of hospital onset E.Coli-AMR cases to prevent the complications like death at hospital level, they are not proactive enough to bring down the raising cases in hospitals. This is because community onset E.Coli-AMR cases have been overlooked – it is the community onset cases that spread the infections within the country, which eventually increases hospitalization, hospital onset cases and serious complications. Hence, it is necessary for the policies to be proactive addressing the issue at community level (upstream) which in turn can reduce the hospital cases. This paper hypothesize and validates a holistic model that captures the interconnections within and between hospital and community sectors using a revised Systems Thinking / System Dynamics (ST/SD) modelling methodology. The revised method helped in reducing the number of parameters in the hypothesized model from 22 to 4 and identifying key leverage points in the system. A preliminary model-based policy evaluation has been performed targeting the leverage points that would bring down the E.Coli-AMR cases