Generic and configurable diagnosis function based on production data stored in Manufacturing Execution System

Ngoc Hoang TRAN, Sébastien HENRY, and Eric ZAMAÏ
Submission Type: 
Full Paper
phmec_16_057.pdf727.98 KBJune 10, 2016 - 3:11am

Nowadays in a highly competitive environment, to control their production systems increasingly complex with wide range products, manufacturers must have precise knowledge about their production systems via data analysis tools. The information system like Manufacturing Execution System (MES) are designed for this requires. MES likes a bridge that’s between the ERP and local control, integrates several functions to control a production system based today on a standard data base (ISA95). The MES solutions provided by editor offer generic functions such as recipe management, execution production, traceability or performance analysis. MES solutions collect and record a growing number of production data especially with the development of unitary traceability. However, the used of these data is often limited into the calculations of key performance indicators, as Overall Equipment Effectiveness (OEE). The evolution analysis of these indicators has just to be performed by the users (operator, team leader, production manager, direction) based on their knowledge on the production system and especially their expertise. In the complex and high variability context, the required time for these “manuals” analyses become incompatible with production requirements. To support human in their phases of analysis of performance indicators drifts, we propose a new generic and configurable diagnosis function based on production data collected and stored by Manufacturing Execution System.
Our objective is to provide maximum information of the origins of a degradation indicator OEE (below the threshold value) and to help making the best decision for all user categories (operator, leader team, supervisor, direction). This analysis will show the macroscopic results corresponding to the specific MES level in the instable production context such as the variation of series, workload or qualification of operators. In our framework, we consider three indicator components of OEE associated with three failures modes: Availability, performance and quality drift who can be calculated on different time periods (hour, day, week, month, and year). In our specific context of MES unique database, probabilistic approaches like Bayesian Network (BN) are well suited techniques to analyze the large amount of production data and can be performed without understanding the underlying structure of a production system. However, a difficulty of BN approaches is to identify set of characteristic variables and the graphical structure during learning phase by exploiting the MES standard data model and the historical data production system.
Therefore, the first step of our approach is an analysis of MES standard data model who defines all the production data on the basis of the ISA 95. We have also proposed a set of potential causes that may impact the successful completion of production operations such as the stress operator, quality material, equipment or changement recipe… This phase has not only presented to identify the observable causes via standard data model of MES and the new potential causes who need to be considered in MES level (such as equipment health, human experience…) but also all the parameters that allow to characterize these causes from the database of MES on the different time horizons. In summary, the BN model has structured on these proposed variables will not only localize quickly and accurately the potential causes of threes indicators drift in MES level but also allow evaluating the suspect level of causes for supporting the maintenance plan by learning the behavior of the production system.

Publication Year: 
Publication Volume: 
Publication Control Number: 
Page Count: 
Submission Keywords: 
Manufacturing Execution System; Fault Diagnosis; Bayesian Network; OEE; decision support
Submission Topic Areas: 
Model-based methods for fault detection, diagnostics, and prognosis
Submitted by: 

follow us

PHM Society on Facebook Follow PHM Society on Twitter PHM Society on LinkedIn PHM Society RSS News Feed