The purpose of the article is to create a predictive analytics simulation model to help managers anticipate manufacturing issues. It integrates specifically the involvement of human resources in the manufacturing systems. The predictive analytics simulation model also includes the main existing interactions between the operators and the manufacturing system.
The semiconductor manufacturing process begins with raw wafers which are thin disks of silicon or gallium arsenide. Several thousand identical chips can be produced on a single wafer. A wafer fabrication plant is called a fab.
A fab is made up of different work areas (or workshops). A workshop contains different workstations. Workstations are closely linked logically or with their location. A workstation is a collection of machines that offer similar processing capabilities. The wafers are collected in an entity called a lot. A lot contains a maximum of 25 wafers (sometimes 50 wafers depending on the diameter).
In the semiconductor manufacturing system, operators are human resources who help to transport lots between different workshops, load, and unload machines, etc.
The research team opted for AnyLogic to build an agent-based simulation model of a manufacturing system. The model consists of the following agents: Entry and Exit of a workshop, the Stocker, the Production Interface, the Machine, and the Operator.
The agent-based approach allows for modeling the system as a population of individual objects (agents or active entities) that communicate with each other and follow a series of rules to achieve their specific objectives. Communication is provided by a set of messages which includes the agent sending a message, the agent receiving the message, and the content of the message.
In the modeled manufacturing system, a workshop is part of the fab. It contains a collection of machines that offer similar capabilities. The workshops remain connected through the different lot flows. The team start by building a generic simulation model able to reproduce the functional behaviour of a workshop. They apply it to a pilot workshop and, depending on the results, they will generalize it to other workshops. To describe in the model the operation of a workshop, the team follow the progress of the lot within the workshop.
The paper presents a predictive analytics simulation model to help semiconductor manufacturing companies have a better understanding of the production flows. An agent-based simulation of the manufacturing system has been built with an emphasis on operator management. Also, the main components of the model have described their behavior.
The manufacturing simulation measures the ability of operators to carry out production and gives the manager an overview of what could be done within a certain time.
This type of model is complex to develop. It took time to get the right data and validate the model. Also, the model and data should be updated regularly. An expert capable of applying the required modifications can measure the divergence. Therefore, the use of this predictive analytics and decision-making tool requires dedicated people.