Abstracts Track 2022


Area 1 - Intelligent Control Systems and Optimization

Nr: 3
Title:

Heuristic Forecast based Upgrade of PID Control, Taking into Consideration Overheating of Rooms

Authors:

Gerfried H. Cebrat

Abstract: The common approach for the control of the room temperature is the placement of such a control unit in a reference room which is used and not exposed to extremes in heating demand and solar gain. In case of major solar gains, thermostats in exposed rooms will close earlier, but not preventing overheating, especially if the release of the heat is via floor or wall heating in a concrete bed. The project ERANet RegSys EPC4SES investigates Model Predictive Control and as such has high interest in optimizing control approaches. Thus the following setting is proposed. The reference room could be a room without much solar irradiation into it. On the other hand, the thermostats in rooms with higher solar irradiation are controlled via µC connected to weather forecast. The PID control then can be amended by an input signal derived for the irradiation forecast. If this forecast is given for several orientations, then CPU load can be kept lower. The paper examines such an approach and calculates potential savings stemming from the avoidance of overheating of the rooms.

Nr: 4
Title:

Model-based Inverse Reinforcement Learning Control of a Batch Crystallization Process

Authors:

Brahim Benyahia, Paul D. Anandan and Chris D. Rielly

Abstract: Pharmaceutical manufacturing relies heavily on crystallization as the main purification technology. The critical quality attributes of the pharmaceutical products, such as drug safety and efficacy, are significantly determined by the performance of the control system or strategy being implemented during the crystallization stage. In addition, downstream processing, such as filtration and drying, is extremely sensitive to small deviations in crystal product quality. The development of effective control of the critical crystal properties such as: size and shape distribution, purity, and polymorphism is challenging due to the complex underlying phenomena and multiple sources of uncertainties. Despite the significant progress made to date in crystallization process control, there is still an increasing demand for more robust and versatile control strategies motivated by more systematic digital quality control, the current Quality-by-Design paradigms, and the resurgence of artificial intelligence. This work presents a novel implementation of Inverse Reinforcement Learning (IRL) approach in the case of a batch cooling crystallization. Here, the Reinforcement Learning (RL) agent observes the expert’s optimal control policies and attempts to mimic its performance. In essence, an Apprenticeship Learning (AL) setup was developed where the expert demonstrates the control task to the IRL agent to help attain effective control performance when compared to the expert. This is achieved through repeated execution of “exploitation policies” that simply maximizes the rewards over the consecutive IRL training episodes. The cooling crystallization of paracetamol is used as a case study and both proportional integral derivative (PID) and Model Predictive Control (MPC) strategies were considered as expert systems. A model based IRL technique is implemented to achieve effective trajectory tracking of the optimal quality profiles which include process temperature, supersaturation, and mean crystal size considered here as the critical quality attribute. The performance of the trained IRL agent was validated against the PID and MPC and tested in presence of noisy measurements and model uncertainties.

Nr: 5
Title:

Adjustment of Vehicle Headlamps with Compensatory Elements using Digital Twin

Authors:

Petr Beremlijski and Michaela Bailová

Abstract: The contributed talk presents a novel strategy for reducing the geometric error of a particular product - a vehicle headlamp equipped with a set of calibration screws. The calibration screws are used to adjust the optimum position of the headlamp. The automated product adjustment procedure was designed to find an optimal configuration for a combination of calibration screws, i.e. a position of the screws in which the distances between the test points and their prescribed positions are minimal and geometric error is minimized. Our strategy involves solving two sub-problems: the design of a digital twin for a headlamp and optimization using calibration screws. We propose a general method for designing and implementing the digital twin, which can be used to minimize overall geometric error. The main idea of developing a digital twin for a headlamp (the first sub-problem) is based on the assumption that the product is a rigid body. We formulated optimal product adjustment (the second sub-problem) as minimizing the locally Lipschitz continuous cost function, which in our case is continuously differentiable and subject to inequality constraints, i.e. it is written as a problem of constrained minimization. We used the gradient method and the Broyden–Fletcher–Goldfarb–Shanno (BFGS) method to solve the optimization problem. In this talk, we present numerical experiments illustrating the solution of both subproblems and the use of our approach. The proposed digital twin allows finding optimal compensatory element settings, leading to minimal total geometric error. Products are automatically adjusted by these settings during the manufacturing process. The novel strategy allows producing parts with approximately 30% more precise tolerances than previously used approaches. Our deployed implementation in C# language running on a regular industrial computer requires only a few seconds (1s to 5s), and the entire machine cycle takes approximately 40 seconds. It means that one machine can adjust more than 2,000 parts per day. In a previous approach, one produced part per day was taken from a production line and precisely adjusted manually by the operator using a screwdriver and a coordinate measuring machine measurement. The same calibration screw setting was used for every part that followed that day, so the same geometric error for each piece was expected. The solution presented in this talk is currently applied in the automotive industry and has been used to adjust approximately 200,000 headlamps. The proposed approach was introduced in the paper [1]. References [1] Jaromír Konečný, Michaela Bailová, Petr Beremlijski, Michal Prauzek, Radek Martinek: Adjustment of Products with Compensatory Elements using Digital Twin: Model and Methodology, PLOS ONE (under review).