Figure 4 - uploaded by Bruce D'Ambrosio
Content may be subject to copyright.
Source publication
The application of Bayesian networks for monitoring and diagnosis of a multistage manufacturing process is described. Bayesian network were designed to represent individual parts in-process. These were combined to form a a Bayesian network model of the entire manufacturing process. An efficient procedure is designed for managing the Simulated data...
Citations
... Several applications of BNs in assembly systems were evaluated by [13]. These applications include monitoring/diagnosis in the multistage cap alignment process [14], fixture failure diagnosis [15], a systematic approach for process fault diagnosis based on BN considering incomplete data and varying noise levels. Additional application examples of ML in manufacturing systems include topological data analysis [16], deep learning [17], and genetic algorithms to evaluate form tolerances [18]. ...
Reducing the dimensional variability of the body-in-white (BIW) in automotive manufacturing is perhaps the most difficult quality control problem due to complex interdependencies amongst the multiple assembly stations that a BIW must pass through in a bodyshop. As increasing quantities of dimensional data are generated in factories, manufacturers face the challenge and opportunity to derive value from the data by enabling advanced quality control methods that can realize greater dimensional stability. As the BIW moves through the bodyshop, dimensional deviations propagate and amplify to downstream stations affecting final vehicle fit-and-finish and visible quality aesthetics potentially influencing a customers’ purchase decision. Current BIW quality approaches rely on univariate statistical process control (SPC) charts. With the large amounts of complex data produced, such charts often fail to detect quality patterns that may exist in hyper-dimensional spaces. As a stop-gap measure, manufacturers attempt to remediate quality issues by assigning operators in final vehicle assembly to visually identify and manually fix apparent deviations. This paper illustrates the application of artificial intelligence (AI) to develop a real-time monitoring system that seeks to predict and detect early dimensional quality issues and eliminate the need for costly downstream corrective actions. Moreover, beyond early detection and prediction, the proposed system also facilitates diagnosis of root causes and understanding the true nature of quality issues.
... Bayesian networks (BN) have been used widely in quantifying the uncertainty of manufacturing processes. Wolbrecht et al. [17] used it for monitoring multistage manufacturing process . McNaught and Chan [18] used BN for fault diagnosis and Nannapaneni et al. [19] used BN in welding and molding processes . ...
Recent advancements in predictive machine learning has led to its application in various use cases in manufacturing. Most research focused on maximising predictive accuracy without addressing the uncertainty associated with it. While accuracy is important, focusing primarily on it poses an overfitting danger, exposing manufacturers to risk, ultimately hindering the adoption of these techniques. In this paper, we determine the sources of uncertainty in machine learning and establish the success criteria of a machine learning system to function well under uncertainty in a cyber-physical manufacturing system (CPMS) scenario. Then, we propose a multi-agent system architecture which leverages probabilistic machine learning as a means of achieving such criteria. We propose possible scenarios for which our proposed architecture is useful and discuss future work. Experimentally, we implement Bayesian Neural Networks for multi-tasks classification on a public dataset for the real-time condition monitoring of a hydraulic system and demonstrate the usefulness of the system by evaluating the probability of a prediction being accurate given its uncertainty. We deploy these models using our proposed agent-based framework and integrate web visualisation to demonstrate its real-time feasibility.
... The application of Bayesian networks for fault diagnosis can be found in medical diagnostics (Cruz-Ramírez et al., 2007, Seixas et al., 2014, engineering systems (Sahin et al., 2007, Cai, Liu, andXie, 2016), manufacturing processes (Yang and Lee, 2012, Wolbrecht et al., 2000, Dey and Stori, 2005, and many other fields. ...
Heating, ventilation, and air-conditioning (HVAC) equipment faults and operational errors result in comfort issues and waste of energy in buildings. In order to help the facility managers to identify and fix faults more efficiently, it is essential to have an Automatic Fault Detection and Diagnosis (AFDD) tool, able to automatically detect comfort and energy issues and identify the root faults.Existing AFDD methods mostly focus on equipment-level fault detection and diagnostics. Almost no attention is given to building level fault diagnosis, considering inter-dependency between equipment through the energy distribution chain.This thesis proposes a new building AFDD method based on operation data collected by Building Management Systems (BMS). The method uses Bayesian Network to achieve building-level integrated fault diagnosis and equipment-level data-driven fault detection by information fusion of data collected from different equipment of HVAC systems. An important contribution relates to the use of operating data and learning techniques to automatically tune some parameters of the detection tool.Our methodology is composed of the following two parts:1. A new systematic way of transferring building system topology information and expert knowledge to a Bayesian Network.2. A novel approach for integrating equipment-level fault detection results into a building-level fault diagnosis Bayesian network. We use regression methods for central equipment (e.g. chiller, boiler, and Air Handling Unit), learned from normal operation data collected during a commissioning test. For room equipment, we use probabilistic models of correlations between control and measurement data.Once the fault diagnosis network is set up and all of the evidence is collected, the network is able to calculate the probability of different faults and identify the most probable root faults. We implemented the fault diagnostics Bayesian network on one simulation data set and two real building operation data sets to test the performance of the AFDD method. The results show that the method is able to easily handle large numbers of equipment, and correctly identify root causes with given evidences.Compared to existing AFDD methods, the new method provides the following benefits:1) The modular structure and generalized methodology allow the method to be applied to wide variety of HVAC systems.2) The method connects equipment faults to building comfort symptoms perceivable by occupants.3) The HVAC system is diagnosed as a whole instead of equipment by equipment.4) By connecting comfort set point violation with equipment fault, and tracing root fault of room equipment failure, the total number of alarms is reduced.5) Facility managers can use the tool in an interactive way, thanks to the capability to post evidence in the Bayesian network based on field investigation findings.
... Bayesian networks (BN) have been used widely in quantifying the uncertainty of manufacturing processes. Wolbrecht et al. [17] used it for monitoring multistage manufacturing process. McNaught and Chan [18] used BN for fault diagnosis and Nannapaneni et al. [19] used BN in welding and molding processes. ...
Recent advancement in predictive machine learning has led to its application in various use cases in manufacturing. Most research focused on maximising predictive accuracy without addressing the uncertainty associated with it. While accuracy is important, focusing primarily on it poses an overfitting danger, exposing manufacturers to risk, ultimately hindering the adoption of these techniques. In this paper, we determine the sources of uncertainty in machine learning and establish the success criteria of a machine learning system to function well under uncertainty in a cyber-physical manufacturing system (CPMS) scenario. Then, we propose a multi-agent system architecture which leverages probabilistic machine learning as a means of achieving such criteria. We propose possible scenarios for which our architecture is useful and discuss future work. Experimentally, we implement Bayesian Neural Networks for multi-tasks classification on a public dataset for the real-time condition monitoring of a hydraulic system and demonstrate the usefulness of the system by evaluating the probability of a prediction being accurate given its uncertainty. We deploy these models using our proposed agent-based framework and integrate web visualisation to demonstrate its real-time feasibility.
... Bayesian networks (BN) is one of the most effective theoretical models for decision making, especially for uncertain knowledge reasoning [1]. In recent years, Bayesian networks have been widely used in a variety of domains such as medical diagnosis [2], device fault diagnosis [3], and system modeling in multiple situations [4][5][6]. Bayesian network learning is the fundamental topic in Bayesian network research. ...
Bayesian network structure learning from data has been proved to be a NP-hard (Non-deterministic Polynomial-hard) problem. An effective method of improving the accuracy of Bayesian network structure is using experts’ knowledge instead of only using data. Some experts’ knowledge (named here explicit knowledge) can make the causal relationship between nodes in Bayesian Networks (BN) structure clear, while the others (named here vague knowledge) cannot. In the previous algorithms for BN structure learning, only the explicit knowledge was used, but the vague knowledge, which was ignored, is also valuable and often exists in the real world. Therefore we propose a new method of using more comprehensive experts’ knowledge based on hybrid structure learning algorithm, a kind of two-stage algorithm. Two types of experts’ knowledge are defined and incorporated into the hybrid algorithm. We formulate rules to generate better initial network structure and improve the scoring function. Furthermore, we take expert level difference and opinion conflict into account. Experimental results show that our proposed method can improve the structure learning performance.
... For example, in the automobile body assembly process, optical coordinate measurement systems have facilitated the full measurement of dimensions for ensuring the quality of the final products [2,3]. As a result, an increasing number of researchers have focused on the development of data-driven monitoring systems for manufacturing processes [3][4][5][6][7][8]. A comprehensive review of the current methodologies for quality control in multistage systems is provided in [9]. ...
For implementing data analytic tools in real-world applications, researchers face major challenges such as the complexity of machines or processes, their dynamic operating regimes and the limitations on the availability, sufficiency and quality of the data measured by sensors. The limits on using sensors are often related to the costs associated with them and the inaccessibility of critical locations within machines or processes. Manufacturing processes, as a large group of applications in which data analytics can bring significant value to, are the focus of this study. As the cost of instrumenting the machines in a manufacturing process is significant, an alternative solution which relies solely on product quality measurements is greatly desirable in the manufacturing industry. In this paper, a minimal-sensing framework for machine anomaly detection in multistage manufacturing processes based on product quality measurements is introduced. This framework, which relies on product quality data along with products’ manufacturing routes, allows the detection of variations in the quality of the products and is able to pinpoint the machine which is the cause of anomaly. A moving window is applied to the data, and a statistical metric is extracted by comparing the performance of a machine to its peers. This approach is expanded to work for multistage processes. The proposed method is validated using a dataset from a real-world manufacturing process and additional simulated datasets. Moreover, an alternative approach based on Bayesian Networks is provided and the performance of the two proposed methods is evaluated from an industrial implementation perspective. The results showed that the proposed similarity-based approach was able to successfully identify the root cause of the quality variations and pinpoint the machine that adversely impacted the product quality.
... Modeling and fault detection of these processes normally requires extensive efforts on collecting, handling and analyzing processes and product measurements, which often introduces higher investment in the necessary infrastructure in managing the high-volume multi-dimensional data. Previous research has mostly been focused on developing multi-stage process monitoring methods which necessitate the measurement of product quality at different stages of the process (Asadzadeh & Aghaie, 2012;Shu & Tsung, 2000;Tsung, Li, & Jin, 2006;Wolbrecht, D'Ambrosio, Paasch, & Kirby, 2000;Zhou, Ding, Chen, & Shi, 2003;Zhou, Huang, & Shi, 2003). ...
Data-driven modeling and fault detection of multi-stage manufacturing processes remain challenging due to the increasing complexity of the manufacturing process, the lack of structural data, data multi-dimensionality, and the additional difficulty when dealing with large data sets. The implementation of add-on sensors and establishing data acquisition, transfer, storage and analysis has the potential to facilitate advanced data modeling techniques. However, besides the associated costs, dealing with high-volume multi-dimensional data sets can be a major challenge. This paper presents a novel methodology for early fault identification of multi-stage manufacturing processes using a statistical approach. The major advantage of the proposed methodology is its reliance on only the product quality measurements and basic product manufacturing records, given the presence of peer sets. This leads to a feasible fault identification solution in a sensor-less environment without investing costly data collection systems. The developed methodology transforms the end-of-process quality measurements to a process performance metric based on a density-based statistical approach and a peer-to-peer comparison of the machines at one stage of the process. This approach allows one to be more proactive and identify the problematic machines that could be affecting product quality. A case study in an actual multi-stage manufacturing process is used to demonstrate the effectiveness of the developed methodology.
... A system framework and a weighted-coupled networkbased dynamic quality control method are proposed to improve the machining errors of one key feature in production process [46,47]. In order to monitoring and estimating the status of quality features, a general approach for quality monitoring and diagnosing in MMPs utilizing Bayesian networks is presented [48]. Furthermore, a qual-ity control fractal network is established for extended enterprises by considering the constraint relationship among nodes [49]. ...
... Finding: In the 12 papers on this topic (8% of the selected papers) [38][39][40][41][42][43][44][45][46][47][48][49], the network models of multistage machining processes are proposed by combining the real network characteristics with complex networks theory. The variation propagation and quality control for multistage machining processes based on a number of measurement indicators are analyzed. ...
In recent years, with the rapid development of manufacturing, information, and management technology, advanced manufacturing systems (AMSs) have become increasingly more and more complex, which hinders the wider applications of many key theories and technologies in AMSs. Fortunately, in the last two decades, some dramatic advances have been made in the field of statistical physics theories, along with the extensive applications of complex network. It has provided an alternative approach to analyze AMSs. Many recent studies have focused on the theory of complex networks to describe and solve complicated manufacturing problems. Based on a great number of relevant publications, this paper presents an up-to-date literature review with the identified outstanding research issues, future trends and directions. Three critical issues are summarized after this investigation: (a) the focused areas of AMSs that have deployed the theory of complex networks, (b) the addressed issues and the corresponding approaches, and (c) the limitations and directions of the existing works.
... The traditional approach often relies on extensive domain knowledge of the process to identifying the critical features [4]. A Bayesian network approach was proposed for manufacturing process monitoring [5]. Each part was represented by the Bayesian network model, and parts models were combined to form a process model. ...
The implementation of advanced technologies in manufacturing has created large amounts of data. The data can be utilized to create predictive models for quality control, which allows manufacturers to produce higher quality products at a lower cost. Bosch has provided a large-scale data set of a production line and hosted a challenge on Kaggle aiming to predict the manufacturing failures using the anonymized features. We proposed a two-stage method first to cluster the data into groups based on the manufacturing process and then use supervised learning to predict the failed product in each cluster. This approach reduces the sparsity of the data set. Various algorithms were compared. The random forest algorithm achieved the highest performance score and was chosen as the final model.
... To identify the required manufacturing-domain knowledge, we studied several research efforts on the applications of data analytics (DA) to manufacturing processes. In [12], the authors apply analytics to detect faults in the alignment of a cap to the base part of a product. In [13], [14], and [15], the authors predict product quality using three DA algorithms: Bayesian networks (BNs), linear regression, and NNs. ...
To employ data analytics effectively and efficiently on manufacturing systems, engineers and data scientists need to collaborate closely to bring their domain knowledge together. In this paper, we introduce a domain-specific modeling approach to integrate a manufacturing system model with advanced analytics, in particular neural networks, to model predictions. Our approach combines a set of meta-models and transformation rules based on the domain knowledge of manufacturing engineers and data scientists. Our approach uses a model of a manufacturing process and its associated data as inputs, and generates a trained neural network model as an output to predict a quantity of interest. This paper presents the domain-specific knowledge that the approach should employ, the formal workflow of the approach, and a milling process use case to illustrate the proposed approach. We also discuss potential extensions of the approach.
