Data-driven order correlation pattern and storage location assignment in robotic mobile fulfillment and process automation system
Introduction
Digital transformation fosters technological evolution, and operational efficiencies have gained the attention of industrial practitioners and researchers. Robotic process automation (RPA), equipped by the software robot through an expert system, can be replaced for the routine, manually repetitive tasks, highly constructed tasks, including automated case creation for the clients [1], automated email query processing [1], process mining [2], and inventory management [3]. Table 1 shows the applications of RPA in different industries, including accounting, banking and financial services, healthcare, human resource, insurance, and government. The rules-based business process in RPA is applied in a software robot, which can operate in a cloud environment or non-cloud environment. The practitioners can gain advantages by adopting RPA to enhance the overall process effectiveness and reduce the workers’ insignificant workload. With the knowledge, workers can spend more time on other value creation tasks without carrying out the repetitive procedures and tasks that are usually done by workers. Customer expectation and satisfaction can be enhanced by RPA because of a higher process efficiency accuracy, regulatory compliance, reliability and a lower scope of human error [4], [5], [6]. The Intelligent Automation (IA), a concept combining RPA, Artificial Intelligence (AI) and Soft Computing (SC) can expand the intelligence by data-driven approach [7], [8]. Technological capabilities and readiness can be improved in the context of IA. By combing the concepts of AI into RPA, for example, RPA can observe the workers solving techniques and handle non-standard or non-repeated cases and provide adaptive decision making.
The emerging cloud-based Cyber-Physical Systems (CPS) framework in the warehouse can capture the operational activities and consolidate the data into a centralized data sink. The adoption of cloud-based CPS can reduce local computational power [24], [25], [26], [27], [28], [29], [30], [31]. The workers need to manually control, monitor, and consolidate all the operations without adopting CPS in the warehouse [24], [32], [33], [34], [35], [36], [37], [38]. With the aid of the Internet of Things (IoT), sensors, the physical layer of CPS can consolidate the environment information and assist with a digitalized environment for nearly real-time control and planning in Robotic Mobile Fulfillment System (RMFS) [39], [40]. For example, customer service integrated with a chatbot for using RPA to help answer simple customer questions with less manpower. RPA can also replace the manual repetitive process, including manually hand-writing for inputting the customer order into the warehouse management system. RPA can also be applied through the cloud-based framework. Boysen, et al. [41] summarized an overview of warehousing systems suited for e-commerce according to the level of automation. In RMFS, a higher level of automation should be adopted for fulfilling the e-commerce customers’ requirements.
The Amazon KIVA system, the first version of RMFS, has been developed for carrying out the e-commerce customer’s order requirement [42], [43], [44], [45], [46], [47], [48], [49]. The major difference between traditional automated guided vehicles and RMFS is that RMFS includes the autonomous mobile robot, moveable rack, and real-time swarm robots control center. de Koster, et al. [50] illustrated that the cost of order picking is estimated to be as much as 55% of the total warehouse operating costs. Under the e-commerce economy, the demand fluctuation and the seasonal effect have more impact on the warehouse operations. One of the reasons why priority orders have increased is because of the introduction of virtual shops for enhancing customers’ loyalty. The product attributes are similar under different online shops. Therefore, order priorities are considered as a value-added service in warehouses. For reducing the order dispatching time and enhancing the efficiency, the layout in the warehouse can be considered as different zones to relieve the storage location assignment problem compared to random assignment. The zone picking policy can increase warehouse efficiency under the synchronized zone order picking system [51], [52], [53], [54]. Data mining for multiple clustering methods will be adopted in this paper for a large order data record in RMFS. The purpose of clustering different zones is to provide greater throughput and reduce the total traveling costs in RMFS. After clustering the zones, the classification methods are used for historical data analysis. The system design is based on human expert judgment and learned from the expert. The major objective is to classify all the orders by the RPA itself without any human expert involvement. The semi-supervised and unsupervised clustering algorithms for zone clustering are compared based on the total traveling costs.
Zone clustering, customer classification, and storage location assignment problems were considered by scholars to reduce warehouse operation costs. For the customer classification problem, Gong, et al. [55] considered the order classes featured by several robots transporting moveable racks to order pickers with developing the Markov models, but only classified two classes. Gharehgozli and Zaerpour [56] extended the asymmetric traveling salesman model by adding general precedence constraints for customers' order priorities. Xie, et al. [57] proposed a new mixed integer model considering splitting orders for enhancing the overall throughput. For the storage location assignment problem, Kim, et al. [58] considered an item assignment problem in the RMFS in order to maximize the sum of similarity values of items in each rack but the method has not been extended for combining the zone clustering in the RMFS. Li, et al. [59] concentrated on the the energy-consumption-aware evaluation method for the storage location assignment policy. For the zone clustering problem, Roy, Nigam, de Koster, Adan and Resing [51] considered the zone assignment strategies under the single-deep scenario. Lamballais, et al. [60] optimized the number of moveable racks per SKU, the ratio of the number of pick stations to replenishment stations, and the replenishment level per rack in the RMFS by introducing a new type of semi-open queueing network. Keung, Lee, Ji and Ng [39], Keung, et al. [61], and Lee, Lin, Ng, Lv and Tai [40] introduced a CPS framework for collision avoidance, conflict resolution, and charging schedule, but the framework has not been considered the RPA. Multiple works could be done by a software robot to reduce human errors. A data-driven approach should also be considered to reduce the total operating costs in RMFS, especially for storage location assignment problem and order classification problem. It is easier to discover patterns and conduct prediction based on the data mining approach.
The contribution of this article has been outlined below. First, a cyber-physical system based on robotic mobile fulfillment and process automation system architecture is proposed. To the best of the authors’ knowledge, no scholars are combining the concepts of RPA and CPS. The CPS structure integrates into the physical layer and cyber layer [62], [63], [64]. Cloud-based RPA can enhance the scalability and mobility of the company. The AS-IS and TO-BE process [1] for generating the order in RMFS are also described in Section 2. Second, Mobile Robots (MRs) cost planning under a multi-deep scenario is considered. Normally, the researchers conduct the RMFS researches under the single-deep scenario [56], [59], [60]. In this paper, the layout includes single-deep and multi-deep operations. The multi-deep scenarios require swarm robot strategies for order fulfillment [39]. The further modified A* algorithm is adopted for calculating the total traveling cost of each moveable rack in Section 3 [65]. Third, considering the results of Section 3 and the historical data of the total number of orders for each moveable rack, multiple clustering algorithms are used for zone clustering in RMFS in Section 4. For assigning the tasks based on multiple zones, it can provide a greater throughput [51]. Therefore, nine common clustering algorithms are applied for the RMFS’s zone clustering. Besides the queuing methods, this paper mainly focuses on the effect of zone clustering for storage location assignment problems and order classification problems by measuring the total traveling costs. The results from Section 4 are considered as 9 scenarios for data-driven order classification. Six common classification algorithms are applied in Section 5, including statistical inference. The concluding remarks are raised in Section 6.
Section snippets
Cyber-physical system-based robotic mobile fulfilment and process automation system
Keung, Lee, Ji and Ng [39] proposed and developed a cloud-based CPS 5C level architecture for improving the operation’s efficiency and effectiveness in the RMFS. A cloud-based CPS is also adopted in this paper and further developed in the RPA scenario, which is shown in Fig. 1. For the model layer, IoT is adopted in RMFS, which includes the elements of MRs, mobile moveable rack, workstation, and charging station. A multi-sensor system is developed in the RMFS. For transferring the information
Steps for single-deep and multi-deep scenario in robotic mobile fulfillment system
Under the RPA for RMFS, a set of tasks is assigned for a set of . QR code is adopted for identifying the location for the racks and the mobile robot. In Fig. 4, the light blue cells denote a single-deep moveable rack and the red ones refer to a double-deep moveable rack, which requires multi-robot collaboration and moves the blocked rack to the temporary zone (TZ). The green ones refer to the TZ. The ones colored in dark red refer to a three-layers multi-deep moveable rack, which
Data collection for zone clustering
Petersen Charles [54] illustrated that the storage has a significant effect on picking zone configuration. Jane and Laih [53] and Kuo, Kuo, Chen and Zulvia [52] further analyzed the clustering algorithm to item assignment in a synchronized zone order picking system, which adopted heuristic or meta-heuristic methods to solve the NP-hard model. However, the scenario in RMFS and the traditional synchronized zone order picking system is different. Normally, in RMFS, one to two workstations are
Data collection for storage location assignment classification
Since there are no public datasets available for the RMFS storage location assignment problem, we adopted the case company RMFS’s historical data also for conducting the storage location assignment classification. The dataset is the same as the zone clustering dataset but considers more information for the classification problem in RMFS. Table 8 lists examples of a historical order dataset for storage location assignment classification. The historical data of assigning the customers’ order was
Concluding remarks
The managerial implications are introduced by justifying how the well-entrenched conceptualization of RPA can be developed further, towards better reducing human errors, with the help of IoT and cloud-based CPS. We propose a framework through cloud-based RPA in CPS and combine a data-driven clustering and classification methods for storage location assignment problems that authors came up through theatrical and practical contemplation. This paper aims to explore the potential behind the RPA and
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgment
This work was supported by the Department of Industrial and Systems Engineering, The Hong Kong Polytechnic University, Hong Kong (RK2F). Our gratitude is also extended to the Research Committee and the Department of Industrial and Systems Engineering, The Hong Kong Polytechnic University, Hong Kong and The Innovation and Technology Commission, The Government of the Hong Kong SAR, Hong Kong for support of this project (PRP/002/19FX/K.ZM31). The authors would like to express their appreciation to
References (119)
A systematic literature review on intelligent automation: Aligning concepts from theory, practice, and future perspectives
Adv. Eng. Inf.
(2021)A human-centred approach based on functional near-infrared spectroscopy for adaptive decision-making in the air traffic control environment: A case study
Adv. Eng. Inf.
(2021)- et al.
A framework for data-driven informatization of the construction company
Adv. Eng. Inf.
(2019) Data-driven operational risk analysis in E-Commerce Logistics
Adv. Eng. Inf.
(2019)Robotic Process Automation: Contemporary themes and challenges
Comput. Ind.
(2020)Edge-cloud orchestration driven industrial smart product-service systems solution design based on CPS and IIoT
Adv. Eng. Inf.
(2019)Online monitoring and collaborative scheduling method for wheelset cyber-physical production system: A wheelset manufacturing system case study from a Chinese high-speed train enterprise
Adv. Eng. Inf.
(2021)A survey of smart product-service systems: Key aspects, challenges and future perspectives
Adv. Eng. Inf.
(2019)Virtual engineering of cyber-physical automation systems: The case of control logic
Adv. Eng. Inf.
(2019)Cloud based cyber-physical systems: Network evaluation study
Adv. Eng. Inf.
(2019)