Toward Data-Driven and Multi-Scale Modeling for Material Flow Simulation: Characteristic Analysis of Modeling Methods

Figures & data

Table 1. An example of activities in a queue model.

Activity	Activity behavior	Required information
Dispatching	Select jobs to be processed next from jobs in a queue.	Job dispatching rule
Resource assignment	Select resources (e.g., machine, worker) to be used for the next jobs.	Resource assignment rule
Setup	Determine whether setup operations are required based on the difference of product type between the previous job and successive job processed by the resources, then occupy the resources for a predefined time of setup operations if needed.	Setup operation occurrence conditions and setup time for each condition
Machining	Occupy the resources for a predefined time of machining operations.	Machining time for each product type
Routing	Move jobs to the queue of the next process.	Process route for each product type

Table 2. Classification of modeling methods.

Modeling method	Model scale	Model type	Definition
Activity WBM	Activity	White box	Describe the behavior of activity explicitly and deductively based on background knowledge.
Activity BBM		Black box	Build an ML model that implicitly express the input–output relationship of the activity from activity log.
Process BBM	Process	White box	Build a queue model in which all activities are expressed by activity WBM and the connections between activities is also explicitly described.
Process BBM		Black box	Build an ML model that directly predict the process output from input.
Process GBM		Gray box	Build a queue model in which sum/all activities are expressed by activity BBM and the connections between activities is explicitly described.
System WBM	System	White box	Build a model in which all processes are expressed by process WBM (used in general simulation software).
System BBM		Black box	Build an ML model that directly predict the system output from input.
System GBM-1		Gray box	Build a model in which some/all processes are expressed by process GBM/BBM.
System GBM-2			Build a model in which any subsystem, which is composed with some processes, in the entire system is expressed by system BBM.

Figure 1. Schematic view of the proposed modeling approach.

Figure 2. Experimental conditions: target production system.

Table 3. Accuracy of each model for System a and System B.

Target system	Num. of machines	R2 of job lead time [-]
		GBM-A	GBM-B	BBM
System A	1	1.00	1.00	0.97
	3	1.00	0.62	0.97
	5	1.00	0.52	0.97
System B	1	0.79	1.00	0.73
	3	0.79	0.77	0.75
	5	0.66	0.69	0.71

Table 4. Accuracy when machining time parameters have $\pm$20% error.

Target system	Num. of machines	R2 of job lead time [-]
		GBM-A	GBM-B
System A	1	0.94	N/A
	3	0.92	N/A
	5	0.89	N/A
System B	1	N/A	0.99
	3	N/A	0.72
	5	N/A	0.53

Figure 3. Generalization ability analysis of each model for System A.

Figure 4. Generalization ability analysis of each model for System B.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author, S.N., upon reasonable request.