Batch Chemical Plants Research Articles

A recurrent reinforcement learning strategy for optimal scheduling of partially observable job-shop and flow-shop batch chemical plants under uncertainty

This study presents a methodology that makes use of Deep Recurrent Q-Learning to develop an agent that acts as an online scheduler for flow-shop or job-shop batch plants with zero-wait restriction under uncertainty. The environment is assumed to be partially observable, i.e., it does not follow the Markov property and information has to be gathered from previous time intervals. The processing times of the machines are unknown to the agent whereas production demand realizations are provided during the operation and not known a priori. The agent aims to complete the demands and to minimize the makespan of the process. Moreover, the agent should avoid violation of constraints associated with product allocation, time horizon, and storage capacity. Three case studies featuring two job-shops and one flow-shop are presented to show the benefits of this framework with environments where the information is limited. Results showed that the agents can generate schedules considering the uncertain parameters of the system while aiming to reduce the makespan of the process. Tests on the agents resulted in small errors in the decision-making process (less than 2%) thus demonstrating that DRQN can serve as a reliable tool for online scheduling subject to uncertainty in partially observable job and flow-shops.

A time-bucket MILP formulation for optimal lot-sizing and scheduling of real-world chemical batch plants

We propose a new time-bucket MILP model for lot-sizing and scheduling problems arising in chemical batch plants. The main idea behind time-bucket models is to partition time into fixed-length macroperiods and flexible length microperiods, that lie within the macroperiods. We show that the time-bucket model benefits from advantages of both continuous and discrete time representations. It allows to include important real-world constraints, can be solved with moderate computational effort, and thus promotes MILP for large-scale, industrial problems. We investigate the scalability of the model and apply it to a formulation and filling process from an industrial agrochemical production with 7 formulation lines, intermediate buffer tanks, and 7 filling lines. We optimize a one month period with 50 intermediates, and 83 finished products. A comparison of the MILP solution to a discrete event simulation solution shows that 17% of production capacity can be freed up and significant improvement in on-time delivery.

An insight-based approach for batch water network with flexible production scheduling framework

Batch chemical plants play an important role in the chemical and pharmacy industry. Efficient water management is becoming increasingly important in batch production systems because of the stricter environmental regulations and rising costs of water utilization activities. In this work, an insight-based method is proposed to effectively utilize the water in batch chemical plant with flexible scheduling framework. First, Pinch Analysis for continuous process is extended to targeting the freshwater requirement of batch processes. Second, a matrix of source-sink matches is constructed based on insight from Pinch Analysis. Some heuristic rules are defined to select the matches of water sinks and sources in order to reduce the connections and meet the time constraints. Thirdly, initial water network is designed based on the chosen matches and the timing of batch operations is determined based on the water reuse. Four illustrative examples adopted from the literature are investigated based on the proposed method. Results show that flexible product scheduling is useful to reduce freshwater consumption and batch operation time. The proposed method could be used to assist wastewater recovery in flexible batch plant.

Stochastic programming-based mathematical model and solution strategy for chemical production scheduling with processing time uncertainty

Scheduling of multipurpose batch chemical plant is always affected by uncertain factors, including processing time of tasks. When the processing time deviates from its nominal value, the task sequence and executing time based upon the original schedule may become suboptimal or even infeasible. To address this issue, an optimization model based on stochastic programming is proposed for the short-term scheduling of multipurpose batch chemical plant, by introducing task sequence variables and new logical constraints relating multiple binary variables. Additionally, a network-based decomposition solution strategy, accounting for different situations of profits and shared units, is proposed to solve the large-scale problems, which has been shown to provide high quality solutions while consuming substantially less solution time than solving the entire process directly.

Machine Learning and Inverse Optimization for Estimation of Weighting Factors in Multi-Objective Production Scheduling Problems

In recent years, scheduling optimization has been utilized in production systems. To construct a suitable mathematical model of a production scheduling problem, modeling techniques that can automatically select an appropriate objective function from historical data are necessary. This paper presents two methods to estimate weighting factors of the objective function in the scheduling problem from historical data, given the information of operation time and setup costs. We propose a machine learning-based method, and an inverse optimization-based method using the input/output data of the scheduling problems when the weighting factors of the objective function are unknown. These two methods are applied to a multi-objective parallel machine scheduling problem and a real-world chemical batch plant scheduling problem. The results of the estimation accuracy evaluation show that the proposed methods for estimating the weighting factors of the objective function are effective.

Optimal scheduling of a multiproduct batch chemical plant with preemptive changeover tasks

Most short-term scheduling formulations in Process Systems Engineering assume a time horizon without interruptions. However, local working patterns may prevent execution of activities at nights or during weekends. One such problem is reported in Egli and Rippin (1986) and deals with a multiproduct batch chemical plant featuring recipes with stable and unstable intermediates, shared resources with limited availability of electricity and steam, non-instantaneous transfer times, and the possibility of interrupting changeover tasks when encountering a pre-defined break period. The objective function balances sequence-dependent changeovers with product holding costs. In this paper, we model these features with a Resource-Task Network (RTN) formulation based on a discrete-time representation. After solving the original problem to optimality for savings of 5%, we evaluate the alternative of forcing complete execution of changeover tasks before or after a break, for a couple more scenarios. The results show that preemption is always beneficial, with improvements ranging from 0.57% to 36.4%.

Integration of Scheduling and Control for Chemical Batch Plants under Stochastic Uncertainty: A Back-Off Approach

Integration of Scheduling and Control for Chemical Batch Plants under Stochastic Uncertainty: A Back-Off Approach

Integration of Scheduling and Control Under Stochastic Parametric Uncertainty with Varying Unit Operation Times for Chemical Batch Plants: A Back-Off Approach

Abstract A new back-off methodology is presented in this work as an approach for solving MIDO formulations arising for the optimal scheduling and control of flow-shop batch plants under stochastic parametric uncertainty. The proposed algorithm decomposes the MIDO problem into a scheduling problem, a dynamic optimization problem and a unit time operation minimization problem. These problems are solved iteratively using back-off terms. Parametric uncertainty is modeled using statistical distribution functions and are embedded in the algorithm to ensure dynamic feasibility of the optimal control actions under stochastic realizations in those parameters. The proposed algorithm identifies scheduling and control decisions offline. To exemplify this methodology, the integration of scheduling and control of a flow-shop batch plant is considered. The results show that unit operation times chosen from optimization are better suited to accommodate stochastic parametric uncertainty while the control actions enforce process operational and product quality constraints at reasonable costs.

Sensitivity Analysis of Genetic Algorithms for Optimization of Operating Conditions for Protein Production Plants

This work introduces genetic algorithms, which will be investigated and used to perform sensitivity analysis. Genetic algorithms are part of a collection of stochastic optimization algorithms based on the concepts of biological evolutionary theory. This research deals with the problem of the search for optimal investment cost of multiproduct batch chemical plants found in a chemical engineering process with uncertain demand. The aim of this work is to minimize the investment cost and find out the number and size of parallel equipment units in each stage. For this purpose, it is proposed to solve the problem by using Genetics Algorithms (GAs). This GAs consider an effective mixed continuous discrete coding method with a four point crossover operator, which take into account, the uncertainty on the demand using Gaussian process modeling. Experiments indicated that relatively good results could be obtained using 4-point crossover with an applied rate of 0.7 and mutation rate ranged between 0.01 and 0.09 promised to give best performance. The results (number and size of equipment, investment cost, production time (Hi), CPU time and Idle times in plant) obtained by GAs are much more rapidly than mixed integer linear programming. This methodology can help the decision makers and constitutes a very promising framework for finding a set of “good solutions”.

Discrete-continuous dynamic simulation of plantwide batch process systems in MATLAB

Batch chemical and biochemical plants play an important role in the process industries. They are characterised by hybrid (continuous & discrete) dynamics as well as complex sequences and decision logic in the case of shared resources. This is challenging from a modelling and simulation perspective, both in terms of numerical algorithms as well as implementability and scalability/maintainability within software environments. In this work it is shown that it is possible to model complex plantwide batch processes at reasonably high performance, accuracy, and practicability in MATLAB/Simulink using the StateFlow toolbox. To this end, useful implementation guidelines are presented, and a complex example batch process is modelled. As focus lies on the implementability of complex batch control logic, the model is limited to mass balances. The simulation results are evaluated and carefully visualised, indicating the MATLAB's capabilities for analysis of such systems.

Decomposition approaches for the design and scheduling of multiproduct multistage batch plants with parallel lines

Multiproduct chemical batch plants, often equipped with parallel production lines, are facilities where similar products are produced by sharing available equipment and time. Hence, when designing such plants, the operational use should be considered simultaneously. In this paper, the design and scheduling problem of a batch plant with parallel lines is addressed, using the Resource-Task Network formulation. The proposed mixed-integer linear programming model aims at minimising total costs through the optimisation of the number of lines to install, their design and the number, size and timings of the batches of every product on the installed lines. Besides an integral (exact) approach, we applied three decomposition approaches to tackle the computational complexity associated with this problem: line decomposition, bottleneck decomposition and a combination of both. Through several examples, it is shown that all three decomposition approaches reduce the computation times significantly compared to the integral approach, while obtaining (near) optimal solutions.

Gate-to-Gate Energy Consumption in Chemical Batch Plants: Statistical Models Based on Reaction Synthesis Type

Energy consumption in the chemical industry is an important operating cost and environmental impact factor and reducing it is also explicitly mentioned as one of the key principles of green chemistry. Energy consumption has thus been included in diverse process design and evaluation tools as a key metric. However, measurements of energy consumption at the process equipment level are scarce, especially in fine chemical production typically performed in multiproduct and multipurpose batch plants. In this work, we present a shortcut approach based on statistical models, such as probability density functions (PDF) and classification trees, for estimating steam consumption which typically represents the highest energy utility consumption in batch plants. The output of these models is in the form of intervals derived from PDF interquartile ranges and as classes derived from the classification trees, respectively. The validation results (i.e., goodness of fit, cross validation, and case studies) show that the mo...

Design of a chemical batch plant with parallel production lines: Plant configuration and cost effectiveness

Abstract We present a model for the design of multiproduct sequential batch plants extended with parallel production lines. This model is meant to support strategic capacity decisions and is formulated as a MILP model. First, we introduce the concept of parallel production lines as a new design option into existing plant design models. Then, we optimise the number of production lines, their design and the product assignment to the installed lines by minimising capital costs of the equipment. Furthermore, we extend the objective function with startup and contamination costs and study the influence of these costs on the chosen plant design options. We find the presence of parallel production lines beneficial as not all products have to share all equipment anymore. Moreover, we show that the incorporation of operating costs affects volume-wise asset utilisation per batch. An example to illustrate the applicability of our model is presented.

Timed-automata based method for synthesizing diagnostic tests in batch processes

Hardware failures are inevitable but random events in the useful life of any batch chemical plant. If these incidents are not efficiently diagnosed, the consequences can be very serious. In general, two design measures may be implemented offline to enhance the overall diagnostic performance, i.e., installing sensors and/or stipulating test plans for online implementations. Since the former has already been studied extensively, the present study focuses only upon the latter. In a recent work, Kang and Chang (2014) proposed an effective method to conjecture diagnostic tests using the untimed automata. However, due to a lack of time-tracking mechanisms, the failure-induced behaviours cannot always be characterized adequately with such models. A systematic procedure-synthesis strategy is therefore developed in the present study by making use of the timed automata and the model-checking capabilities of existing software, e.g., UPPAAL (Behrmann et al., 2006). All component models are first constructed, and all possible fault propagation scenarios and their observable event traces (OETs) are next enumerated exhaustively. The optimal test plan for every OET can then be established by generating the supervisory controller to improve diagnostic resolution. Extensive case studies have also been carried out in this work to confirm the validity and effectiveness of the proposed approach.

Automata generated test plans for fault diagnosis in sequential material- and energy-transfer operations

Hardware failures are inevitable random events that occur in the operation life of a batch chemical plant. Based on the piping and instrumentation diagram (P&ID) of the given process and the sequential function chart (SFC) of its normal operating procedure, a system automaton and the corresponding “diagnoser” can be built to identify all observable fault propagation traces and, also, their root cause(s). Since the fault origin(s) of a trace may not be unique, there is a need to develop a nonconventional means to further enhance diagnostic performance. For this purpose, a novel approach is proposed in this study to synthesize the test plan of every undiagnosable trace on the basis of discrete-event system (DES) theory. In particular, all components at the failure-induced initial states and the required control specifications are first modeled systematically with automata and, then, an optimal supervisor (test plan) can be assembled accordingly so as to achieve the operation goal of differentiating the fault origins as much as possible. This proposed strategy has been tested successfully in a series of examples and the results of selected case studies are reported in this paper.

Industrial Documentation-Based Approach for Modeling the Process Steam Consumption in Chemical Batch Plants

One of the key principles of green chemistry is the minimization of energy consumption in the chemical industry. Besides continuous production processes, which have traditionally been the focus of ...

複数ジョブをもつ離散型並列生産システムのための動作モデルの提案

A formal model named Expanded-Timed-State-Chart (ETSC) for expressing the complex behavior of a discrete production system (DPS), where multiple jobs are simultaneously carried out in parallel, was proposed by expanding Timed-State-Chart model. Making use of the model, event-driven simulation of DPS in formal fashion can conveniently be carried out, and so the verification method based on Model-Checking becomes applicable to DPS. In order to illustrate the usefulness of the model, computational experiments were carried out for verifying the validity of operations given to a target DPS, that is a multi-purpose chemical batch plant given by Japan Batch Forum. Some practical assumptions to reduce the “reachable-tree” to a restricted one have been introduced here in order to avoid the difficulty of combinatorial explosion, that intrinsically appears in the generation of Kripke-Structure of original DES.

Model-Based Identification and Analysis of the Energy Saving Potential in Batch Chemical Processes

Optimization of energy consumption for reducing the relevant costs and environmental impacts is constantly gaining attention in chemical batch production. Existing methodologies focus on heat integration considering scheduling constraints and typically result in trade-offs between capital investment and operational costs. However, in multipurpose batch plants, even the allocation of energy flows and the consistent operation according to production recipes pose a great challenge due to batch-to-batch variability and lack of energy utility consumption meters. This paper utilizes a bottom-up modeling approach for energy utility consumption and proposes a method for model-based identification of the energy saving potential in chemical batch plants. The bottom-up models can accurately track the energy utility consumption at various production levels and are used as soft sensors for energy efficiency analysis studies. In this context, a set of energy key performance indicators (EKPIs) is proposed for quantifying efficiency in energy use, and an energy saving potential index (ESPI) based on historical plant performance serves as a shortcut method in the case of missing or inaccurate production recipes. The methodology is applied to an industrial multipurpose batch plant for specialty chemicals, exemplifying the obtained efficiency results and targeting energy saving potential for steam consumption.

Estimation and Analysis of Energy Utilities Consumption in Batch Chemical Industry through Thermal Losses Modeling

A systematic approach for the estimation of energy utility consumption in chemical batch plants is presented and validated. This approach is based on bottom-up modeling of energy use and its conversion in usual energy carriers like steam, cooling water and brine, including estimation of thermal losses. The modeling involves a detailed energy balance of each unit operation using dynamic plant data. Thermal losses are determined using an empirical parametric equation for each equipment and utility. The method was applied to one monoproduct and one multiproduct building of a chemical batch plant comprising 20 main equipments (reactor vessels, heat exchangers and dryers) for the production of 5 specialty chemicals and intermediates over a period of 2 months. For fitting the parameters of the empirical thermal losses models it was crucial to determine the real energy consumption for the investigated equipments. In the lack of installed flowmeters it was tested whether this task can be performed through calibration of the valves controlling the utility distribution based on valve opening. A direct and an indirect calibration method were applied both providing satisfactory level of accuracy. Validation of the bottom-up models was performed at different aggregation levels, including equipment specific unit operations, production lines and overall building energy consumption. The results indicated that for all types of energy utility, equipment and unit operation the consumption could be estimated with a relative error between 5% and 35% depending on the aggregation level. These results can motivate plant managers and process engineers to consider model-based estimation and analysis of energy utilities for monitoring and optimization purposes in industrial practice.

An automaton-based approach to evaluate and improve online diagnosis schemes for multi-failure scenarios in batch chemical processes

Online diagnosis has been considered as an important measure for improving operational safety in many batch chemical plants. Specifically, the state transition behaviors of all hardware items (components) in the given batch process and their failure mechanisms are modeled systematically with automata. The system model is then assembled by connecting the component models on the basis of a generic hierarchical structure. A “diagnoser” can be constructed accordingly for the purpose of determining various qualitative and quantitative performance indices. Guided by these indices, two performance enhancement approaches can be effectively applied: (1) installing additional sensors which are not included in the piping and instrumentation diagram (P&ID) and (2) executing extra operation steps which are not specified in the sequential function chart (SFC). Three examples are presented in this paper to demonstrate the feasibility of the proposed approach.