Available online at www.sciencedirect.com
ScienceDirect Procedia CIRP 61 (2017) 116 – 121
The 24th CIRP Conference on Life Cycle Engineering
Life cycle simulation of mechanical parts with part agents considering user behavior Yumihito Yokokia,*, Hiroyuki Hiraokaa a
Depertment of Precision Mechanics, Chuo university, 1-13-27 Kasuga, Bunkyoku, Tokyo, Japan
* Corresponding author. Tel.: +81-3-3817-1841; fax: +81-3-3817-1820. E-mail address:
[email protected]
Abstract
To realize the effective reuse of mechanical parts for the development of a sustainable society, it is essential to effectively manage individual parts over their entire life cycle. For this purpose, we are developing a part agent system using network agents. A part agent manages all information about a part throughout its life cycle and predicts possible states of the part in the near future in order to generate appropriate recommendations for their maintenance. User behavior is important for this prediction because, in the life cycle of a part, its states are affected by the user behavior. The behavior of users is modeled using prospect theory in order to simulate its nature so as to avoid the risk of failure. This paper compares the decisions for maintenance actions in the simulations with and without applying prospect theory. The initial results are presented in a life cycle simulation of the developed part agents. © 2017 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license © 2017 The Authors. Published by Elsevier B.V. (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the scientific committee of the 24th CIRP Conference on Life Cycle Engineering. Peer-review under responsibility of the scientific committee of the 24th CIRP Conference on Life Cycle Engineering
Keywords: Prospect theory; Life cycle; User behavior.
1. Introduction To realize the effective reuse of mechanical parts, which is a competent measure of the development of a sustainable society [1][2], it is essential to manage the individual parts over their entire life cycle. Manufacturers face difficulties in predicting the quality and quantity of the used parts that are necessary to perform reuse-based production, owing to the uncontrollable and unpredictable variation in user behavior. Product users also experience difficulties in carrying out appropriate maintenance on many and various parts in their products. Based on these considerations, we propose a scheme whereby a part "manages" itself and supports user maintenance activities, and we verify it using a life cycle simulation. Life cycle simulation has been recognized as an effective tool to design product life cycles [3]. To realize this scheme, network agents that are programmed to follow their real-life counterpart parts throughout their life cycle are being developed. These network agents are referred to as
“part agents” [4] and provide users with appropriate advice on the reuse of their part and promote the circulation of reused parts. Previous works have proposed methods of user support when using a part agent [4][5][6]; however, the user may not accept the proposed actions recommended by the part agent based on his or her preferences and requirements. For example, even when the probability of failure of a product is low, the user might want to replace it with a new product after seriously considering the risk of failure. Therefore, it is necessary to develop a user model and method that conforms to the behaviors of real users. In addition, the occurrence of failure and deterioration should be taken into consideration in the prediction made by part agents and in the life cycle simulation [7]. To deal with the replacement and purchase of parts, the exchange of information between the current part and other parts is required in order to relate the life cycles of different parts. In this paper, we describe a method developed to predict the future state of a part with a life cycle connected
2212-8271 © 2017 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the scientific committee of the 24th CIRP Conference on Life Cycle Engineering doi:10.1016/j.procir.2016.11.235
Yumihito Yokoki and Hiroyuki Hiraoka / Procedia CIRP 61 (2017) 116 – 121
to that of other parts. Another aim of this paper is to deal with the effect of user behavior on the maintenance of the part. For this purpose, prospect theory is applied to part agents for representing user behavior, and the occurrence of failure of the parts is also simulated. This paper describes a part agent’s proposal of maintenance actions that takes into consideration the use behavior. A mechanism is proposed to incorporate the occurrence of part failure in the life cycle simulation. The results of the simulation with and without the application of prospect theory are then compared. This paper consists of the following sections. First, the concept of a part agent is described in section 2. Thereafter, the mechanism of a part agent that provides advice on the reuse of a part is explained in section 3. Next, the representations of the life cycle of the parts and user behavior are described in section 4. In section 5, methods to represent exchange of parts in the life cycle simulation and a life cycle simulation including the replacement of parts and failure of parts are explained. In section 6, the results of the simulation to evaluate the effect of user behavior are presented. Finally, the paper is concluded in section 7.
117
data, as well as the management and control functions of the product. Communication is established using information agents that are subordinate agents generated by the part agents [5]. 3. Creation of advice by part agent Fig. 2 shows a framework for a part agent to advise the user of the necessary maintenance actions based on the life cycle model of its corresponding part. At each time step, the part agent predicts the possible states of the part in the near future and evaluates those options in order to provide the user with appropriate advice [6].
2. Part Agent system A part agent manages all information regarding its corresponding part throughout its life cycle. The proposal assumes the spread of networks and high-precision radio frequency identifier (RFID) technology [8]. A part agent is generated during the manufacturing phase of the core parts where an RFID tag is attached to each part. The part agent identifies the ID of the RFID tag during the life cycle of the part, tracking the part through the network. We chose an RFID tag for identification because RFIDs have a higher resistance to smudge or discoloration than printed bar codes and will last for the entire life of the part.
Fig. 1. Conceptual scheme of the part agent
Fig. 1 shows the conceptual scheme of the part agent. The part agent communicates with various functions within the network and collects the information required to manage its corresponding part, such as product design information, predicted deterioration of parts, logistic information, or market information. It also communicates with local functions on-site, such as sensory functions that detect the state of the part, the storage functions for individual part
Fig. 2. Framework of part agent for advice generation
A part agent generates advice in the following order. First, the part agent expands the life cycle of the part by using the current status of the part and its evaluated values based on a deterioration model. All possible candidate paths in the life cycle are derived and estimated using a user model. The part agent obtains information on the candidate replacement parts from other agents in the market and uses it in the prediction. Based on the estimation, an appropriate action for the part is selected and recommended to the user. Details on the expansion and estimation of the life cycle of a part are described in section 5. A variety of information is required for part agents to evaluate appropriate maintenance actions. This includes the benefit acquired from the part, the required cost for the part, and the environmental load generated in relation with the part. As these values change primarily owing to deterioration, a deterioration model that represents how the mechanical performance of the part deteriorates is required. To deal with failures of parts, we assume that the probability of failure of a part depends on the deterioration of the part. A part agent predicts the probability of failure from a deterioration model, and estimates the probability with which stages, such as use, repair, and others, will be selected in the future. Prediction of the probability of failure and the derivation of path probability are described further in section 5 and 6.
118
Yumihito Yokoki and Hiroyuki Hiraoka / Procedia CIRP 61 (2017) 116 – 121
4. Prediction of user behavior
details).These functions are implemented at the selection of the next stage to realize user behavior.
4.1. Evaluation of life cycle To generate advice to the user, a part agent predicts future states of the corresponding part based on its life cycle model. Fig. 3 shows a simple example of the life cycle model used in this paper. The life cycle of a part consists of life cycle stages and life cycle paths that connect them. The circles in Fig. 3 represent the life cycle stages, which include produce, sell, use, repair, and dispose. The arrows indicate the life cycle paths.
Fig. 3. Simple life cycle model
The part agent expands this life cycle of the part and represents possible changes in its life cycle over time. Fig. 4 shows an example of an expanded life cycle of the part starting from the "use" stage.
5. Life cycle simulation including the replacement of parts 5.1. Exchange of parts Exchange of parts such as the replacement of a part with a reused part and the purchase of a part after the disposal of the current part is important to promote the reuse of parts. However, the life cycle shown in Fig. 3 represents the life cycle of a single part and cannot represent the exchange of parts. In other words, the life cycle of a part should be extended to include the viewpoint of the user. For this purpose, we introduce a stage "replace" and a path from "dispose" to "sell and buy" into the life cycle model as shown in Fig. 5. It should be noted that parts are exchanged in these newly added stages, and paths that are marked with asterisks indicate that the life cycle of a part is connected to that of another part as shown by the chained arrows in Fig. 6.
Fig. 5. A life cycle model that incorporates the exchange of parts
Fig. 4. Expanded life cycle model
Each stage in the expanded life cycle has properties such as benefit, cost, and environmental load. Each expanded life cycle path has a different probability for the part to take that particular path. These properties and probabilities are calculated using the life cycle simulation based on the deterioration model. Part agents evaluate each path to a possible next stage by estimating the expected values taking into consideration the values of the following stages and probabilities of the paths connecting to them. The details of this estimation are included in section 5. 4.2. User behavior The part agent selects the path that brings the highest expected value and advises its user to take that path. The decision is made based on the difference between the gain i.e. the estimated benefit, and the loss i.e. the estimated cost and environment load. The part agent selects the candidate stage with the largest positive difference and recommends it to the user. However, users tend to avoid risks more than the exact estimation. Kahneman summarizes these tendencies of user behavior using two functions that are the value function and the weighting function in prospect theory [9]. The value function represents a mental model on how people evaluate gain and loss. The weighting function represents the nonlinearity of preferences (see Appendix A for more
Fig. 6. Transfer between life cycles of different parts
The life cycle is expanded as shown in Fig. 7. An asterisk indicates an exchange of parts. In the "replace" stage, the reused parts in a market will be evaluated and selected for use in the next stage. The reused part is evaluated by expansion of the life cycle as same as the current part. When a part is disposed of, a new part is introduced and evaluated in the same manner. In the figure, the black circles represent life cycle stages of other parts such as part A and part B. Each part agent corresponding to the part that not only is being used by a user but also exists in a market expands the life cycle of its part. Every agent of the part in a market expands its life cycle, respectively, as shown in Fig. 8, and it offers information based on other agents' demand. Thus, part agent selects the path including the exchange of parts that leads to the stage with the best expectation.
119
Yumihito Yokoki and Hiroyuki Hiraoka / Procedia CIRP 61 (2017) 116 – 121
ܸܧൌ ቌ ோ௨௧
ܸכ
ௌ௧௨௧
ෑ
ܲቍ
(1)
௧௨௧
Here, ܸܧis the expected value of a candidate in the next stage considering the future stages, ܸ is the sum of the property values for the stages in a route, and ܲ is the accumulated probability of the paths in the route. For example, in Fig. 9, the expected property value of Stage1 is calculated using equation (2). Fig. 7. Expanded life cycle model visible to users
ܸܧሺܵͳ݁݃ܽݐሻ ൌ ሺܸͳ ܸʹ ܸͷሻ כሺʹ כ ʹͳͷሻ ሺܸͳ ܸʹ ܸሻ כሺʹ כ ʹͳሻ ሺܸͳ ܸʹ ܸሻ כሺʹ כ ʹͳሻ ሺܸͳ ܸ͵ ڮሻ כሺ כ ͵ͳǥ ሻ ڮ (2)
The benefit, cost, and environmental load of the part are the properties predicted based on the deterioration of the part. In this case, the benefit is defined as a gain, and the cost is defined as a loss. For simulating user behavior, we have applied the value function ݒሺሻ and the weighting function ݓሺሻof prospect theory to the calculation of the expected values as shown in equation (3). Fig. 8. Expansion of the life cycles by the agents in a market
ܸܧൌ ቌ ோ௨௧
ௌ௧௨௧
ݒሺܸሻ כ
ෑ
ݓሺܲሻቍ
(3)
௧௨௧
5.2. Estimation of the life cycle of the part [6] 5.3. Prediction of part failure by part agent A part agent applies the prediction of part failure to the estimation of the life cycle of a part. The probability for the "use" stage and the "replace" stage is calculated from the probability of part failure as described in section 6. We define it as the probability that the user chooses the path. 6. Life cycle simulation 6.1. Conditions of the simulation
Fig. 9. Process of accumulating the values of the life cycle
Fig. 9 describes how a part agent selects the next stage using the expanded life cycle of a part. The figure shows an example situation in which the current life cycle stage (shown to the left) is a "Use" stage and the possible candidate stages in the next time step are Stage1, Stage1', and Stage1". A circle denotes an expanded life cycle stage with its property value ܸͳ, ܸʹǡ and ܸ͵. An arrow denotes an expanded life cycle path with the probability ʹͳ, ͵ͳǡ and ͳͶ . To evaluate each possible candidate stage, the expected value of a property is calculated, taking into consideration the series of paths in the future. A series of stages connected with the paths is defined as a "route." The property values for the next stages and their probabilities are collected for all possible routes that could occur in the future. The expectation is then calculated for each route by multiplying the sum of the property values and the product of probabilities, as shown in equation (1).
We performed the life-cycle simulation for machine parts such as the parts of a computer, like HDDs to which "decomposition" and "replacement" can be applied. This system was developed using Java and Eclipse, operating on a Windows OS. To perform a simulation to estimate the future states of a part, we made a simple assumption that a part deteriorates with its operational time. Our deterioration model for the part is shown in Fig 10. The mechanical performance of the part decreased with its operational time and was recovered when the part was repaired. In this simulation, the mechanical performance decreased by 5% of its initial value for each step of use and recovered up to 40% of its initial performance on repair. We acknowledge that deterioration and failure have more complex and stochastic behavior [10], but we use this simple model as a first step. The properties of parts were updated in the stages for each time step as shown in Table 1. P denotes the current performance of the part and ܲ௫ denotes its initial performance when a new product is created. It should be noted that Platter denotes the performance of the part exchanged with the current part. It is assumed that the cost and environmental load depend on P and ܲ௫ [5].
120
Yumihito Yokoki and Hiroyuki Hiraoka / Procedia CIRP 61 (2017) 116 – 121
Fig. 10. Deterioration model Table 1. Properties of stages Produce Sell and Buy
Cost
Environmental load 0
Performance
Benefit
ܲ௫
0
0
ܲ௫
0
ܲ௫
0.1×P 0.2×ܲ௫ 㸫0.1×P 0.2×ܲ௫
Use
P㸫0.05×ܲ௫
Repair
P㸩0.3×ܲ௫
0
0.2×ܲ௫ 㸫0.1×P 0.3×ܲ௫
Replace
ܲ௧௧
P
ܲ௧௧ െ ܲ
ͲǤͳ ൈ ሺܲ௧௧ െ ܲሻ
Dispose
0
0
0.2×ܲ௫
ܲ௫
Use
Repair
Replace
Dispose
Pr
0.2
Pd
P
Table 2. Probabilities of paths To From
Produce
Produce Sell and Buy Use
Sell and Buy
1.0 1.0 Pu
Repair
benefit against cost and environment load acquired at the stage. Fig. 11 shows the variation of TPI with time. The result of the simulation when considering user behavior is shown in red and that without considering user behavior is shown in blue. Values obtained by the simulations are compared in Table 3. Results without considering user behavior show a larger number of "use" and "repair" stages whereas results considering user behavior show a larger number of "replace" stages. The accumulated TPI value is 177.7 for results with user behavior and 183.4 without user behavior. As users tend to avoid the risk of failure, they select replacement of parts before the probability of failure rises. This frequent replacement decreases the duration for "use", which in turn lowers the TPI value. The results show the effects of user behavior. However, certain issues still remain, such as the simple deterministic nature of the deterioration model, which must be improved. With these experimental conditions, when prospect theory is applied, replacement is chosen prior to repair. This is because a part agent judges that it is more cost-effective to perform part replacement frequently rather than to fix the part and use it for a long time.
1.0
Dispose
1.0
We also employ simple models for probability of part failure and path based on the deterioration of part. Table 2 shows the probabilities of the life cycle paths. Pr, Pd, and Pu represent the probabilities that the part is repaired, disposed of, and used without repair, respectively. We assume that the part in use is replaced with a constant probability of 0.2. fp represents the probability of failure of the part. These probabilities are calculated using equations (4)–(7). ܲ௫ െ ݁ܿ݊ܽ݉ݎ݂ݎ݁ ܲ௫ ܲ ݎൌ ݂ ܲ݀ ൌ ݂ ܲ ݑൌ ͳǤͲ െ ܲ ݎെ ܲ݀ െ ͲǤʹ
݂ ൌ
(4) (5) (6) (7)
Part failure occurs at the "use" stage with probability fp. If part failure occurs, the user cannot select the "use" stage in the next step. The part agent advises the user to select the "repair" stage or the "disposal" stage in the next step. At the "repair" stage, the failure is fixed and its performance is recovered as shown in Table 1. To decide the best stage from candidate stages, the value INDEX is calculated using equation (8), ܺܧܦܰܫൌ ݐ݂ܾ݅݁݊݁ܧെ ݐݏܿܧ
(8)
where Ebenefit and Ecost are expected values of the stage calculated from the benefit and cost, respectively. These expected values are obtained using equation (1) or (3) depending on whether the user behavior is considered (equation (3)) or not (equation (1)). 6.2. Simulation results with and without considering user behavior To evaluate the simulation results, we use the total performance index or TPI [11], which represents the relative
Fig. 11. Variation of TPI in the lifecycle simulation of parts Table 3. Comparison of resultant values for simulation with and without considering user behavior The average number of events Failure of parts
Without prospect theory 2.8
With prospect theory 2.9 20.5
Use
21.1
Repair
1.3
0
Replace
2
3.8
Dispose
2.8
2.9
Sell and buy
2.8
2.8
7. Conclusion In this paper, we propose the generation of maintenance actions with the help of a part agent, while considering user behavior as well as the occurrence of part failures. In addition, the exchange of parts in their life cycle is considered for the generation of part agent advice. Life cycle simulations of parts with part agents are performed and its results show the effect of user behavior based on prospect theory. However, there are still some unresolved issues. As the simulation of this paper deals with a single user, "the difference for every user" has not been considered, but we will deal with the difference of users by performing a simulation with two or more users in the future. Although a linear deterministic model is used in this paper for
121
Yumihito Yokoki and Hiroyuki Hiraoka / Procedia CIRP 61 (2017) 116 – 121
simplification, we recognize the significance of using more practical models that include nonlinearity and stochastic nature. Acknowledgements
probability of an event. It means that if the probability of events increases to more than 35%, a human feels that the event occurs at a lower probability than the original probability, and if it is less than 35%, a human feels that the event occurs at a higher probability than the original probability.
This work was supported by JSPS KAKENHI Grant Number 15K05772. Appendix A. Prospect theory Prospect theory [10] is a theory of behavioral economics published by D. Kahneman and A. Tversky that describes the way people choose between probabilistic alternatives that involve risk. It consists of two characteristic equations explained below. Fig. A1. Value function
A.1.Value function The value function represents the mental value for gain or loss. In prospect theory, if we define ݒas the mental value that a human being feels they will gain or lose ݔ, ݒis concave above the reference point ݔൌ ͲሺݒԢԢሺݔሻ Ͳǡ ݔ Ͳሻ and convex below the reference point ሺݒԢԢሺݔሻ Ͳǡ ݔ൏ Ͳሻ , where ݒԢԢሺݔሻ denotes the second-order differential of ݒሺݔሻ . Furthermore, ݒas the mental value is steeper for losses than for gains (ݒԢሺݔȁ ݔ Ͳሻ ݒԢሺݔȁ ݔ൏ Ͳሻ, where ݒԢ is the derivative of ݒሺݔሻ). The first two conditions reflect the principle of diminishing sensitivity: the impact of a change diminishes with distance from the reference point. The last condition implies the principle of loss aversion [10]. From these features, the value function ݒሺݔሻ given in equation (A1) is denoted by a graph, as shown in Fig. A1. ݔఈ ሺ ݔ Ͳሻ (A1) ݒሺݔሻ ൌ ൜ െߣሺെݔሻఉ ሺ ݔ൏ Ͳሻ where the values of the coefficients are ߙ ൌ ͲǤͺͺǡ ߚ ൌ ͲǤͺͺǡ ߣ ൌ ʹǤʹͷ . These coefficients were calculated through an experiment by A. Tverski and others [12]. This equation is used to evaluate the property values for stages in the life cycle. A.2.Weighting function The weighting function is obtained from the experiment to represent the observed nonlinearity of preferences [9]. The weighting function ݓା ሺሻ and ି ݓሺሻ are given in equations (A2a) and (A2b) for gain and loss, respectively. ఊ ݓା ሺሻ ൌ (A2a) ଵ ሼఊ ሺͳ െ ሻఊ ሽ ൗఊ ି ݓሺሻ ൌ
ఋ ሼఋ ሺͳ െ ሻఋ ሽ
ଵൗ ఋ
(A2b)
Here is the probability of an event, and ߛ ൌ ͲǤͳ and ߜ ൌ ͲǤͻ are coefficients obtained in the experiment [12]. The graph of ݓሺሻ versus the probability is drawn based on these equations, as shown in Fig. A2. The solid line represents the original probability of an event. This graph shows that the shape of the curve changes from concave to convex at approximately 35% of the
Fig. A2. Weighting functions for gains and for losses
References [1]
Hauschild M, Jeswiet J, Alting L. Life cycle assessment to sustainable production: Status and perspectives, CIRP Ann 2005; 54(2):535-555. [2] Tani, T. Product development and recycle system for closed substance cycle society. In Environmentally Conscious Design and Inverse Manufacturing, 1999. Proceedings. EcoDesign'99: First International Symposium, IEEE; p. 294-299 [3] Takata, S., Kimura, T. Life cycle simulation system for life cycle process planning. CIRP Ann-manuf technol 2003; 52(1):37-40. [4] Hiraoka H, Ueno T, Kato K, Ookawa H, Arita M, Nanjo K, Kawaharada H. Part agent advice for promoting reuse of the part based on life cycle information. In: The 20th CIRP Int Conf Life Cycle Eng 2013; 335-340. [5] Nanjo K, Yamamori Y, Yokoki Y, Sakamoto Y, Hiraoka H. Maintenance decisions of part agent based on failure probability of a part using bayesian estimation. In: The 22nd CIRP Conf Life Cycle Engineering, Sydney, 7-9 April 2015. [6] Yokoki Y, Yamamori Y, Hiraoka H. User model in the life cycle simulation of a part based on prospect theory, In: EcoDesign 2015 9th International Symposium on Environmentally Conscious Design and Inverse Manufacturing, 2015; 184-189 [7] Takata, S., Yamada, A., Kohda, T., Asama, H. Life cycle simulation applied to a robot manipulator-An example of aging simulation of manufacturing facilities. CIRP Ann-Manuf Technol, 1998; 47(1):397-400. [8] Borriello G. RFID, Tagging the World, Communications of the ACM 2005; 44(9):34–37 [9] Kahneman D., Tversky A. Prospect theory: An analysis of decision under risk,. Econometrica: Journal of the Econometric Society 1979; 263-291. [10] Herrmann, C., Kara, S., Thiede, S. Dynamic life cycle costing based on lifetime prediction. International Journal of Sustainable Engineering 2011; 4(3): 224-235. [11] Kondo S, Masui K, Mishima N, Matsumoto M. Total performance analysis of product life cycle considering the uncertainties in product-use stage, Adv Life Cycle Eng Sustainable Manufacturing Businesses 2008; 371-376. [12] Tversky A, Kahneman D. Advances in prospect Prospect theoryTheory: Cumulative representation Representation of uncertaintyUncertainty 1992; J Risk Uncertainty 1992; 5(4):297-323.
