Modelling manufacturing processes using Markov chains

Abstract

• This is an Accepted Manuscript of a paper published by CRC Press in INTRODUCTIONModelling manufacturing processes which contain human interactions is difficult and can produce unrealistic views of the process. This is because in many companies the actual manufacturing process that takes place is not as planned when human interaction is involved. Human factors can determine what actually happens, the time it takes and what order it happens in. To produce a more reliable representation of the process more information on what is actually happening is required. This can be found by tracking and recording the process using radiofrequency identification (RFID) tags (Weinstein, 2005). From the data produced from these tags the possible paths which products take in the process can be determined and hence the actual manufacturing process can be defined. Furthermore the data can be used to form Markov chains which can determine what future process routes will look like and the probability of each route. Basing future business simulations on these Markov chains can give a more reliable representation of the business. This reduces the risk of modelling inaccuracies and can help to predict future outcomes and run optimisation more accurately.The research performed here studies a company which refurbishes IT products. The company has a business model of the manufacturing process which it expects the products to follow. The company has tracked their products through the refurbishment process using RFID tags to determine what processes each product undertakes and to allow each object to be kept track of. The RFID tags communicate the process information of each product to a database software using RFID tag readers. The information from these RFID tags is used in the work presented here to form a Markov chain representation of the business process routes.Companies' model manufacturing processes for many reasons, including predicting cost (Rehman et al., 1998), predicting resource and material demand and running optimisation studies.When modelling the data produced from the RFID tags the Markov chain produced gives a large variety of process routes. These are not all true reflections of the routes products take. The data mining process from the RFID tag data is also investigated to allow the development of more precise process models. This process allows thresholds to be set for each route. Hence irregular paths are removed.The Markov chain is necessary to simulate the product flow. The process is simulated using the Markov chain produced from the data and the results can be compared to the process simulated based on previous perceptions of the business process. The results produced will include the time taken for each product to be processed, the cost of the process and the final destination of products. The Markov chain ABSTRACT: Optimizing manufacturing processes with inaccurate models of the process will lead to unreliable results. This can be true when there is a strong human influence on the manufacturing process and man...

Designing a digital twin for micromanufacturing processes

“…48 This fact may increase the complexity as it results in coupling the sensors with the data processing; the concept of prediction, training and classification capabilities of algorithms 49,50 is the other side of the coin. In this case, relevant methods are Markov chains 51,52 and Fuzzy Sets 53 (without excluding the association to physics), while for the case of stochastic nonstationary signals, 54 there is a whole list of metrics that are used. This classification of tools that are useful for digital twins and automated decision making is complemented by other methods, indicatively Taguchi methodology (impact study), 55 Tikhonov regularization for inverse physics models 56,57 and for optimization techniques.…”

“…48 This fact may increase the complexity as it results in coupling the sensors with the data processing; the concept of prediction, training and classification capabilities of algorithms 49,50 is the other side of the coin. In this case, relevant methods are Markov chains 51,52 and Fuzzy Sets 53 (without excluding the association to physics), while for the case of stochastic nonstationary signals, 54 there is a whole list of metrics that are used.…”

Mathematical Modeling of Manufacturing Lines with Distribution by Process: A Markov Chain Approach

“…It is interesting to highlight some efforts made to model mass manufacturing systems with the approach proposed here, see, for example, [26][27][28][29][30]. This proposal is a first attempt to develop a model in process manufacturing systems, and more research is needed in the future to improve the modeling.…”