TY - GEN
T1 - A BAYESIAN NETWORK-BASED CONTROL CHARTING FRAMEWORK FOR IDENTIFYING THE ROOT CAUSES OF PROCESS SHIFTS IN MULTISTAGE MANUFACTURING SYSTEMS
AU - Mondal, Partha Protim
AU - Ferreira, Placid M.
AU - Kapoor, Shiv G.
AU - Bless, Patrick N.
N1 - The authors acknowledge financial support from Intel Inc. through a university research support grant. Ricardo Toro (UIUC) provided technical support with information systems and simulations.
PY - 2024
Y1 - 2024
N2 - Multistage manufacturing systems (MMS) are a pervasive and essential part of high-volume production systems. An MMS is characterized by many interdependent stages, each with multiple process settings, accepting inputs characterized by many features and producing outputs with several features related in a complex manner to the input features and process settings. This makes it challenging to pinpoint specific causes of any quality issues observed in the output of the system. Although conventional control charts are the most widely used tool for quality monitoring in an MMS, they do not provide insights into the root causes of quality departures from normal to help focus corrective actions. Historically, in industrial applications, the responsibility of identifying the underlying root causes of observed quality excursion has rested with human operators, who, in turn, rely on empirical methods to analyze the data and come up with potential causes. Therefore, in this paper, a Bayesian network-based control charting framework is proposed to directly track the current state of process inputs and settings of a multistage system for efficiently identifying the root causes when quality departures are observed in the system’s output. The overall framework consists of two components: a) A Bayesian network (BN) for inferring the states of the process inputs and settings, based on observed system outputs, and b) A control charting procedure for tracking the states of the process inputs and settings to determine if each is in or out-of-control. To illustrate the concepts presented in this paper, a tractable two-stage manufacturing system is used as an example process. The simulation results demonstrate the framework’s practical application, highlighting how it integrates root cause diagnosis into quality monitoring, showcasing a more efficient way of diagnosing root causes in multistage systems.
AB - Multistage manufacturing systems (MMS) are a pervasive and essential part of high-volume production systems. An MMS is characterized by many interdependent stages, each with multiple process settings, accepting inputs characterized by many features and producing outputs with several features related in a complex manner to the input features and process settings. This makes it challenging to pinpoint specific causes of any quality issues observed in the output of the system. Although conventional control charts are the most widely used tool for quality monitoring in an MMS, they do not provide insights into the root causes of quality departures from normal to help focus corrective actions. Historically, in industrial applications, the responsibility of identifying the underlying root causes of observed quality excursion has rested with human operators, who, in turn, rely on empirical methods to analyze the data and come up with potential causes. Therefore, in this paper, a Bayesian network-based control charting framework is proposed to directly track the current state of process inputs and settings of a multistage system for efficiently identifying the root causes when quality departures are observed in the system’s output. The overall framework consists of two components: a) A Bayesian network (BN) for inferring the states of the process inputs and settings, based on observed system outputs, and b) A control charting procedure for tracking the states of the process inputs and settings to determine if each is in or out-of-control. To illustrate the concepts presented in this paper, a tractable two-stage manufacturing system is used as an example process. The simulation results demonstrate the framework’s practical application, highlighting how it integrates root cause diagnosis into quality monitoring, showcasing a more efficient way of diagnosing root causes in multistage systems.
KW - Bayesian networks
KW - machine learning
KW - Multistage manufacturing system
KW - process monitoring
KW - quality control
KW - root cause diagnosis
UR - http://www.scopus.com/inward/record.url?scp=85203706533&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85203706533&partnerID=8YFLogxK
U2 - 10.1115/MSEC2024-130229
DO - 10.1115/MSEC2024-130229
M3 - Conference contribution
AN - SCOPUS:85203706533
T3 - Proceedings of ASME 2024 19th International Manufacturing Science and Engineering Conference, MSEC 2024
BT - Manufacturing Equipment and Automation; Manufacturing Processes; Manufacturing Systems; Nano/Micro/Meso Manufacturing; Quality and Reliability
PB - American Society of Mechanical Engineers (ASME)
T2 - ASME 2024 19th International Manufacturing Science and Engineering Conference, MSEC 2024
Y2 - 17 June 2024 through 21 June 2024
ER -