Adaptive multi-objective service composition reconfiguration approach considering dynamic practical constraints in cloud manufacturing

Abstract

Dynamic uncertainty factors such as equipment faults are common in practically implemented cloud manufacturing (CMfg) environments, often causing the manufacturing service to be invalidated. In that case, efficient reconfiguration of the original service composition under practical constraints is critical; however, existing research scarcely focuses on it. This paper proposes a dynamic service composition reconfiguration model to bridge the gap by considering practical constraints (DSCRPC) in a real-life cloud manufacturing environment. Based on the constraints considered in this study, the DSCRPC model redefines three objectives: time (T*), cost (C*), and product service quality (Q*S*). To optimize the DSCRPC model, this study developed an adaptive multi-population multi-objective whale optimization algorithm (AMPOWOA) based on the Pareto strategy. The algorithm adopts four balancing strategies and adaptively optimizes and adjusts the key parameters under various balancing strategies through well-designed reinforcement learning models. Finally, we conduct numerical experiments and actual application case tests to compare the performances of AMPOWOA and other algorithms (MOWOA, MOHHO, NSGA-II). The results show that DSCRPC can continuously tackle the cloud manufacturing service composition (CMSC) reconfiguration issue with constraints until an order is completed. Moreover, AMPOWOA is superior to the other algorithms optimizing the DSCRPC model. This significantly enhances the robustness of service composition reconfiguration in real-life CMfg.

Introduction

Cloud manufacturing (CMfg) [1], [2], [3], [4], [5] is a novel service-oriented networked manufacturing paradigm, that organizes and optimizes virtual manufacturing resources via a network. CMfg is prevalently employed owing to its advantages of integrating dispersed manufacturing resources and optimizing allocation on demand. Cloud manufacturing service composition(CMSC) [6], [7] has attracted extensive attention as the core issue of CMfg. For instance, Li et al. proposed a multi-objective optimization method for CMSC with an improved particle swarm optimization algorithm [8]. Furthermore, Zhang et al. proposed an extended flower pollination algorithm (FPA), which not only utilizes adaptive parameters but also integrates with a genetic algorithm (GA), to obtain the optimal service composition solution [9]. While scholars currently focus on efficiently matching the optimal cloud manufacturing service chain in an ideal cloud environment [10], [11], few have considered practical constraints caused by manufacturing resources at the executive level. Such unpredictable [12] factors significantly affect the corresponding cloud services, causing the cloud service time to change dynamically or invalidate. Fortunately, with the application of digital twins in the manufacturing industry [13], [14], the refined real-time perception of the manufacturing resources of each layer and service state in the cloud manufacturing process is realized, especially the dynamic factors that seriously affect the cloud service quality. Thus, it is crucial to develop a more practical cloud manufacturing model by combining these dynamic factors [15]. To complete processing tasks smoothly over time, it is currently urgent to reconfigure the service composition efficiently under practical constraints when service exceptions occur.

The CMSC reconfiguration needs to meet practical constraints in the actual service composition execution process, including service exception constraints [16], the strict time constraint of the original CMSC, and service occupation time constraints [17]. In addition, the strong coupling constraint between some existing manufacturing resources is pervasive and non-negligible in the CMSC reconfiguration process. For example, the processing technology standards of different manufacturing equipment series can be coupled. Thus the manufacturing resources selected by two adjacent subtasks may have a strong correlation. Suppose that the upstream coupled manufacturing resource is selected for the previous subtask. In this case, the next adjacent subtask can only select matching downstream coupled manufacturing resources (there may be multiple downstream coupled manufacturing resources) to complete the processing task. Moreover, the tools used in manufacturing resources have a specific service life [17]. Once the life of manufacturing resource tools is over, the manufacturing resource will perform adjustment or reconfiguration, causing the actual processing time to change dynamically [17].

Furthermore, in the cloud manufacturing process [18], [19], manufacturing resources usually have multiple processing speeds. Furthermore, workpieces can be transferred between cities using various transportation methods, including trucks, shipping, and air transport. Therefore, the manufacturing resource processing speed and workpiece transportation significantly affect the service composition reconstruction process. Thus, such processing speed and transportation selection constraints should be considered in the CMSC reconfiguration model [20]. However, to the best of our knowledge, existing research [21], [22] scarcely involves such practical non-negligible constraints to formulate service composition reconfiguration models in real-life cloud manufacturing environments. Therefore, this paper proposes a dynamic service composition reconfiguration model to fill the gaps, considering the aforenoted practical constraints in a real-life cloud manufacturing environment.

The research issue in this study corresponds to a typical combinatorial optimization problem [23], [24], [25], [26]. Many studies have shown that meta-heuristics [27] effectively solve the models and obtain an optimal solution. For example, Que et al. developed an improved adaptive immune genetic algorithm to address the optimal QoS-aware service composition selection in cloud manufacturing [28]. Furthermore, Zhou and Yao [29] proposed a hybrid artificial bee colony algorithm to solve a QoS-based optimal service selection problem. Thus, considering the effectiveness and superiority of meta-heuristics [26], [30], this study solves the presented service composition reconfiguration model with meta-heuristics. In addition, service composition reconfiguration is a multi-objective issue, where objectives may conflict with each other. Nevertheless, analyzing the inter-influence relationship between multiple objectives is crucial to the final decision in real-life cloud manufacturing. Existing service composition reconfiguration methods [17] transform multiple objectives into a single objective by setting weights, resulting in a dilemma that fails to reveal the relationship between multiple objectives. Thus, this study redefines optimization objectives based on the considered practical constraints and develops a Pareto-based method to reveal the relationships between objectives.

In summary, the service composition reconfiguration model considering such real-life dynamic factors has not been reported. In this study, we have developed an improved multi-objective algorithm (AMPOWOA) for the proposed service composition reconfiguration model to narrow the gap between theory and application. The contributions and novelties of this study are divided into the following aspects:

• According to the real-life cloud manufacturing process, eight crucial practical constraints are proposed to formulate a novel service composition reconfiguration model (DSCRPC).

• Based on the considered constraints, the objectives of the DSCRPC were refined, making the model more accurate, refined, and practical.

• The developed multi-objective optimizer (AMPOWOA) includes multiple optimization strategies. In particular, the proposed cluster optimization strategy can adjust the processing mode and the transportation mode adaptively, according to Constraint 8, significantly speeding up the search process for high-quality solutions.

• AMPOWOA, for the first time, combines the multi- population balance strategy and the RL models (Q-learning and SARSA) with eligibility traces to adjust its crucial parameters adaptively and quickly obtain higher-quality solutions.

• The DSCPC model and AMPOWOA can adaptively control the service composition reconfiguration processes in real time under practical constraints until completing the order in an actual cloud manufacturing environment.

The remainder of this paper is organized as follows. Section 2 reviews the related research. Section 3 proposes a dynamic service composition reconfiguration model that considers the practical constraints. In Section 4, we elucidate the developed adaptive multi-population multi-objective whale optimization algorithm (AMPOWOA) based on well-designed strategies. Section 4 presents the numerical experiments and comparisons with state-of-the-art algorithms. Section 5 applies the presented dynamic service composition reconfiguration approach to solve a real-life application case. Finally, Section 6 discusses the conclusions and future work.