Manufacturing Logistics Optimization Using the SPECTER Task Planner: A Shoe Manufacturing Logistics Case Study

Manufacturing Logistics Optimization using the SPECTER Task Planner: A Shoe Manufacturing Logistics Case Study Anatoli A. Tziola and Savvas G. Loizou Department of Mechanical Engineering and Materials Science and Engineering, Cyprus University of Technology, Limassol, CYPRUS, {anatoli.tziola, savvas.loizou}@cut.ac.cy Abstract. This paper presents an application of SPECTER task planner for shoe manufacturing logistics. In this case study we propose an abstract model of the work flow of a shoe manufacturing company taking into account: i) the workflow stages; ii) the time costs (i.e. a machine operation, products time construction, worker transitions between work-cells); and iii) the agents involved in the production line (such as machines, humans, materials, products etc.). Based on the derived abstraction, optimal solutions are provided for utilizing 1, 2 or 3 workers in the production line. The paper validates the modeling power of the SPECTER framework, while demonstrating its potential applications in providing solutions for the industry. Keywords: Task planning, shoe manufacturing logistics, SPECTER. 1 Introduction Solving and automating complex manufacturing logistics tasks in order to produce a variety of products, has been one of the areas of research interest in the field of robotics and in particular their applications to manufacturing logistics. Shoe making is such a labour intensive process that requires many different sizes, styles and materials for shoes. The shoe manufacturing workflow undergoes various and complex steps before a quality product arrives at a customer, making it hard to automate the planning of such a production. Among the major challenges affecting shoe manufacturing are inefficient utilization of resources, including energy, materials and human resources, leading to increase the cost of the final product. The manufacturing of shoes consists of multiple workflows with intermediate stages until a pair of shoes is produced. Some of the workflows are the shoe designing, cutting from leather to textiles, stitching, final assembly and packing. Each workflow is distinguished in more individual steps. This work focuses on the optimization of the final assembly workflow and its individual steps of a shoe manufacturing industry in order to optimize the intra-factory logistics. The challenge is that a single production line consists of many intermediate steps and multiple types of production processes depending on the customer order. 2 Anatoli A. Tziola and Savvas G. Loizou Utilizing the SPECTER task planning framework proposed by the authors in [1], [2], different scenarios are studied, to determine the resources and the time required for shoes production depending on the available human workers in the factory as well as the sequence of actions that need to be performed (i.e. the machines/robots operation sequence, the materials utilization, worker actions). The presented results depend on the assumptions, the accuracy of the data and on the level of abstraction and should be considered as a preliminary result that demonstrates the capabilities of our modeling framework. The rest of the paper is organized as follows: Section 2 presents the description of the workflow, and the workflow abstract model, while section 3 presents the results of 3 different scenarios. Section 4 concludes the paper. 2 Analysis and Modeling 2.1 Description of the Final Assembly Workflow Final assembly is the workflow where upper leather is shaped, lasted and assembled with sole to get final the product i.e. the shoes. The final assembly is performed before the packing procedure. The main techniques of final assembly Back-Part Shaping Unprocessed leather Back-Part Shaped Leather Toe Shaping Unprocessed Leather Toe Shaped Leather Back-Part Shaped Leather Back-Part + Toe Shaped Leather Toe preShaping Toe Shaping Pressing Machine Lasting Preparation Lasting Machine Toe Lasted Leather Back-Part + Toe Lasted Leather Back-Part pre-Shaping Back-Part Shaping Side + Heel Lasting Glue Insertion Side Lasting Machine SideLasted Leather Control Heel Lasting Machine Lasted Upper Leather Painting + Polishing A1-spray Painting B1-spray Painting A2-wax polishing Painted Upper Leather Calibration Machine 2 Calibration Machine 1 Calibration Calibrated Upper Leather Roughing Marking soles Marking soles Cementing shoe + sole Attaching shoe+ sole Pressing Machine Heat Tunnel Soles Brushing Removing Last from shoe Finishing Shoes ready for packing Toe Shaped Leather Packing Workflow Back-Part/Toe Lasting Shoe (upper leather marked with sole) Back-Part + Toe Shaped Leather Fig. 1. Diagram of the final assembly workflow. used in the shoe manufacturing industry, presented in Fig.1 are: Toe Shaping: Unprocessed leather or back-part shaped leather is doubled shaped for preserving the shape and the original appearance of the shoe. Back-Part Shaping: Unprocessed leather or toe shaped leather is doubled shaped for preserving the shape and the original appearance of the shoe. Lasting: The unprocessed leather or back-part shape leather or back-part and toe shaped leather can be processed. In this stage, the upper leather is attached to the bottom of the shoe. Side and Heel Lasting: Setting the final shape of the shoe and holds it in place so the outsole can be permanently attached. Painting and Polishing,Calibration: Calibrating the shoe last. Marking soles: Marking the sole in the upper lasting Manufacturing Logistics Optimization using the SPECTER Task Planner 3 leather, then roughing and cementing shoe with sole. When shoe with sole is attached, the shoe is pressed and then passed through the heat tunnel. Finishing: Removing last from shoe’s inside, then brushing. Finally, the shoe is produced and continues to packing workflow. 2.2 Abstract Model In order to model the workflow shown in Fig. 1 an abstraction is required to cast the problem in the discrete processes domain of the SPECTER framework. Due to space restrictions, an outline of the process abstraction is depicted in Fig. 2. The nodes of the diagram represents the processing steps and buffer zones, whereas the arrows indicate the input and the output products of each process and buffer zone. P1 P2 A3 A1BM3 A3 A7BM7P4 A6 A5 A5BM5 P3 A4 A4 BM4 P2 P1 A2 A1BM2 A2A4 21 21 21 21 34 126 Fig. 2. Workflow abstraction. Numbers above circles indicate duration (sec) Workers and items are modelled as agents expressed as ϵ0-NFAs [1,2]. The detailed representations of each agent are beyond the scope of the current paper. More specifically, the unprocessed leather is modeled as A1; the toe-shaping leather is modeled as A2; the back-part shaping leather is modeled as A3; the back-part and toe-shaping leather is modeled as A4; the lasted leather is modeled as A5; the soles are modeled as A6; and the shoes are modeled as A7. Additionally, 3 workers are considered working in the factory and modeled as A8,A9, A10 respectively. Step P1 represents the back-part shaping, P2 the toe shaping, P3 the backpart and toe lasting processes. The processes of side and heel lasting, painting and polishing, calibration, marking with soles and finishing are grouped in step P4 since no sequence changes are allowed. Moreover, BM 2 represents the buffer zone for items modeled as A2,BM 3 the buffer for A3,BM 4 the buffer for A4, BM 5 the buffer for A5and BM 7 the buffer for A7. The processing steps and the buffer zones are considered as agents locations. The unprocessed leather A1can be converted to the back-part shaping leather A3at P1 and then, A3placed at buffer BM 3. Also, A1can be converted to A2 at P2 and then, A2is placed at buffer BM 2. Item A4can be produced at P1 using item A2or at P2 using item A3. Then, A4is placed at BM 4. Item A5is produced at P3 using item A4. Item A7is produced at P4 using items A5and A6. Finally, item A7is placed at BM 7. 4 Anatoli A. Tziola and Savvas G. Loizou 3 Case Study Results We run 3 scenarios utilizing 1, 2 or 3 workers respectively. We assume that there are unlimited resources of unprocessed leather at P1 and P2; and soles are at P4. For the initial conditions we assume that toe-shaping leather, back-part shaping leather, back-part and toe-shaping leather, lasted leather and shoes have not been produced yet, the workers could be anywhere in the factory. 1st scenario: 1 worker. For the 1st scenario, we considered 8 agents in total; 7 items and 1 worker. The cardinality of the environment’s state space is 14,400 states. The objective is to produce A7and locate it at BM 7 utilizing 1 worker. In the solution computed by SPECTER, the objective is fulfill after 20 steps. In words, (T1) worker goes at work-cell P1, (T2) worker inserts A1in machine at P1, (T3)A3is produced at P1 after 21 seconds, (T4) worker places A3at BM 3, (T5) worker goes at work-cell P2, (T6) worker inserts A1in the machine at P2, (T7)A2is produced at P2 after 21 seconds, (T8) worker places A2at BM 2, (T9) worker inserts A3in machine at P2, (T10)A4is produced at P2 after 21 seconds, (T11) worker places A4at BM 4, (T12) worker goes at work-cell P3, (T13) worker inserts A4in machine at P3, (T14)A5is produced at P3 after 34 seconds, (T15) worker places A5at BM 5, (T16) worker goes at work-cell P4, (T17) worker inserts A5in machine at P4, (T18) worker inserts A6in machine at P4, (T19)A7is produced at P4 after 126 seconds, (T20) worker places A7at BM 7. Thus, the task is fulfilled; shoes produced and places at buffer zone BM 7. The time required to implement the solution is 269 seconds utilizing 1 worker for the whole process. 2nd scenario: 2 workers. For the 2nd scenario, we considered 9 agents in total; 7 items and 2 workers. The cardinality of the environment’s state space is 72,000 states. The objective is to produce A7and locate it at BM 7 utilizing 2 workers. In the solution computed by SPECTER, the objective is fulfill after 16 steps. In words, (T1) worker A8goes at work-cell P2; worker A9goes at work-cell P1, (T2) worker A8inserts A1in machine at P2, while worker A9inserts A1in machine at P1, (T3) after 21 seconds, A2is produced at P2 and A3is produced at P1, (T4) worker A9places A3at BM 3, while worker A8places A2at BM 2, (T5) worker A8inserts A3in the machine at P2, while worker A9inserts A2in the machine at P1, (T6) 2 entities of A4are produced after 21 seconds; 1 entity at P1 and 1 entity at P2, (T7) worker A8places A4entities at BM 4, (T8) worker A8goes at work-cell P3, (T9) worker A8inserts A4in machine at P3, (T10)A5 is produced at P3 after 34 seconds, (T11) worker A8places A5at BM 5, (T12) worker A8goes at work-cell P4, (T13) worker A8inserts A5in machine at P4, (T14) worker A8inserts A6in machine at P4, (T15)A7is produced at P4 after 126 seconds, (T16) worker A8places A7at BM 7. The time required to implement the solution with concurrent execution capability requires 238 seconds. 3rd scenario: 3 workers. For the 3rd scenario, we considered 10 agents in total; 7 items and 3 workers. The cardinality of the environment’s state space is 360,000 states. The objective is to produce A7and locate it at BM 7 utilizing Manufacturing Logistics Optimization using the SPECTER Task Planner 5 3 workers. The solution provided by SPECTER consists of 16 steps from (T1) to (T16) as described in scenario 2 but in this case the solution utilizes worker A9instead of A8and worker A10 instead of A9. The time required to implement the solution with concurrent execution capability requires 238 seconds. Fig. 3 demonstrates the results from the 3 case studies. Fig. 3. Optimal cost (time) for the three case studies 4 Conclusions In this paper a case study for a manufacturing logistics optimization problem is presented, utilizing the SPECTER task planning framework. An abstraction was proposed that modeled key features of the work flow and the resulting model abstraction was implemented on the SPECTER task planner. Three case studies were investigated, with 1, 2 and 3 workers working on the production line. Interestingly a decrease in the production time was observed only when a second worker was added in the workflow but no change was registered when a third worker was added. The results concluded that an increase above the optimal number of workers will not decrease the production time. Further investigations with different machines/robots/buffers and factory layout arrangements are under consideration as future work. 5 Acknowledgements This work was partially supported by European Union’s research and innovation programs under grant agreement 951813 (Better Factory, H2020) and 101092295 (CIRCULOOS, Horizon Europe). The authors would like to acknowledge Mr. Aleksander Zur, CEO of TAPI NERO shoe manufacturing, for the discussions and valuable insights in shoe manufacturing logistics. References 1. A. A. Tziola and S. G. Loizou, “Autonomous task planning for heterogeneous multiagent systems,” IEEE International Conference on Robotics and Automation, 2023. 2. A. A. Tziola and S. G. Loizou, “Autonomous task planning for heterogeneous multiagent systems,” arXiv Preprint arXiv:2209.08611, 2022.