Sensor layout optimization by integrating Bayesian approach to diagnose multi-station assembly processes

Dimensional control has a significant impact on the overall product quality and performance in large and complex multi-station assembly systems. From measurement data, the way to identify root causes for large variation of Key Product Characteristics (KPCs) is one of the most critical research topics in dimensional control. This paper proposes a new approach for multiple fault diagnosis in a multi-station assembly process by integrating multivariate statistical analysis with engineering model. Based on product/process information, by using the state space model, a set of fault patterns for multi-station assembly process are developed, which explicitly represent the relationship between the error sources and KPCs. The vectors of these patterns form an affine system. Afterwards, the Principal Component Analysis (PCA) is applied to conduct orthogonal diagonalization of the measurement data. Thus, the measurement data can be easily projected to the axes of the affine system. Whereby, the significance of each fault pattern shall be estimated accurately. Finally, a few case studies are also provided to validate the proposed methodology.

Fault diagnosis for underdetermined multistage assembly processes via an enhanced Bayesian hierarchical model

In chemical processes, Bayesian network (BN)-based approaches have been extensively applied for process fault diagnosis. Generally, BN is learned using score and search algorithms where search algorithms create candidate networks whose fitness to data is measured by scores. However, existing approaches cannot utilize cyclic loop knowledge while learning BN. Since cyclic loops are prevalent in chemical processes, their unaccountability results in inaccurate BN and reduces diagnosis accuracy. Therefore, for accurate diagnosis, we propose direct transfer entropy (DTE)-based multiblock BN to discover cyclic loops while learning BN. First, the process is segmented into multiple blocks. Next, block-level BNs are learned using DTE-based score and Greedy search. By eliminating the common source variable effect, DTE finds correct causality between process variables and obtains accurate block-level BNs, which are fused to discover significant cyclic loops that were otherwise unachievable when finding BN. The performance of the developed methodology is demonstrated through a benchmark process.

Fault detection and isolation of faults in a multivariate process with Bayesian network

Product dimensional quality is one of the most important topics in machining processes. However, limited research has been done on the dimensional control and root cause identification methodologies in multi-station machining processes. This paper will study the stream of variation (SOV) problem by developing a new type of model — State Space Model — to describe the dimensional deviation of the multistation machining processes. A process level diagnostic methodology is also developed. A case study is presented to illustrate the developed modeling and diagnostics methodology.

Several researchers have studied the use of informational entropy as a surrogate measure for the reliability of water distribution networks. The hypothesis is that the numerical value of network entropy in some way reflects the reliability of water distribution networks, and this appears to be supported by the analysis of some example water distribution networks. However, the precise relationship between the entropy value and some measures of reliability has not been formally established. The primary objective in this paper is to present an alternative methodology to calculate the informational entropy of water distribution networks. This methodology is termed as the Path Entropy Method (PEM), which provides some insights into the entropy of branching-tree networks and maximum-entropy flows of single-source networks. In addition, a quick method of computing the maximum-entropy value of single-source networks is presented and termed as the Simplified Path Entropy Method (SPEM).

A method for obtaining the Bayesian network (BN) representation of a sensor's measurement process is developed so that the problems of sensor fusion and management can be approached from a unified point of view. Uncertainty, reliability, and causal information embedded in the sensor data are used to build the BN model of a sensor. The method is applied to model ground-penetrating radar, electromagnetic induction, and infrared sensors for humanitarian demining. Structural and parameter learning algorithms are employed to encode relationships among mine features, sensor measurements, and environmental conditions in the BN model. Inference is used to estimate target features in the presence of heterogeneous soil and varying environmental conditions. A multisensor fusion technique operating on BN models is developed to exploit the complementarity of the sensor measurements. Through the same approach, a BN classifier is obtained to estimate the target typology. The BN models and classifier also compute so-called confidence levels that quantify the uncertainty associated with the feature estimates and the classification decisions. The effectiveness of the approach is demonstrated by implementing these BN tools for the detection and classification of metal and plastic landmines that are characterized by different shape, size, depth, and metal content. Through BN fusion, the accuracy of the feature estimates is improved by up to 64% with respect to single-sensor measurements, and the number of objects that are both detected and classified is increased by up to 62%.

Linearization is commonly employed in modeling of multi-station manufacturing processes for quality improvement purposes. However, an analysis of the accuracy of the linear models is inevitable before they could be used, because practical manufacturing processes are inherently nonlinear. Such analysis is not straightforward for the high dimensionality of multi-station manufacturing processes and the complicated relationship between process parameters and the process output. This paper presents a nonlinearity analysis method for multi-station assembly processes through an integration of experiment design and data mining. It is found that certain design parameters (e.g., the ratio between locator deviations and the distance between locators) and the number of stations play an important role in the linearization error. Although demonstrated in the specific context of two dimensional multi-station assembly processes, the proposed method can be applied to investigate many characteristics of a broad variety of manufacturing process models.

Many studies have been carried out to investigate the important factors in calculating the realistic entropy amount of water distribution networks, but none of them have considered both mechanical and hydraulic characteristics of the networks. Also, the entropy difference in various networks has not been calculated exactly. Therefore, this study suggested a modified entropy function to calculate the informational entropy of water distribution networks so that the order of demand nodes and entropy difference among various networks could be calculated by taking into account both mechanical and hydraulic characteristics of the network. This modification was performed through defining a coefficient in the entropy function as the amount of outflow at each node to all dissipated power in the network. Hence, a more realistic method for calculating entropy was presented by considering both mechanical and hydraulic characteristics of network while keeping simplicity. The efficiency of the suggested method was evaluated by calculating the entropy of some sample water networks using the modified function.

Robustness of water distribution networks is related to their connectivity and topological structure, which also affect their reliability. Flow entropy, based on Shannon’s informational entropy, has been proposed as a measure of network redundancy and adopted as a proxy of reliability in optimal network design procedures. In this paper, the scaling properties of flow entropy of water distribution networks with their size and other topological metrics are studied. To such aim, flow entropy, maximum flow entropy, link density and average path length have been evaluated for a set of 22 networks, both real and synthetic, with different size and topology. The obtained results led to identify suitable scaling laws of flow entropy and maximum flow entropy with water distribution network size, in the form of power–laws. The obtained relationships allow comparing the flow entropy of water distribution networks with different size, and provide an easy tool to define the maximum achievable entropy of a specific water distribution network. An example of application of the obtained relationships to the design of a water distribution network is provided, showing how, with a constrained multi-objective optimization procedure, a tradeoff between network cost and robustness is easily identified.

<p>An assessment of the rainfall station distribution in the mountainous area of the Regional Autonomous Corporation of Cundinamarca (CAR, for its acronym in Spanish), Colombia, was conducted by applying concepts from information entropy and artificial neural networks (ANNs). This study was divided into two phases: first, a classification of the meteorological stations using two-dimensional self-organizing maps; second, the evaluation of the performance of the ANN by applying concepts of information entropy. Three scenarios were raised for the classification of the meteorological stations by adjusting the number of neurons in the output layer. A high number of neurons in the output layer were obtained, causing the model to over-fit while emphasizing differences amid patterns. When comparing the results of the scenarios, the permanence of certain characteristics and features was found in the system, validating the model classification. Subsequently, the results of the first scenario were used to evaluate the entropy of the historical series. Finally, the results show that the area of study presents a lack of information due to the uncertainty associated with the probabilistic arrangement, which can be corrected with the developed model. Consequently, some recommendations for the redesign of the rainfall are provided. </p>

International audience

Deep learning models are now widely used in decision-making applications. These models must be robust to noise and carefully map to the underlying uncertainty in the data. Standard deterministic neural networks are well known to be poor at providing reliable estimates of uncertainty and often lack the robustness that is required for real-world deployment. In this paper, we work with an application that requires accurate uncertainty estimates in addition to good predictive performance. In particular, we consider the task of detecting a mosquito from its acoustic signature. We use Bayesian neural networks (BNNs) to infer predictive distributions over outputs and incorporate this uncertainty as part of an automatic labelling process. We demonstrate the utility of BNNs by performing the first fully automated data collection procedure to identify acoustic mosquito data on over 1,500 h of unlabelled field data collected with low-cost smartphones in Tanzania. We use uncertainty metrics such as predictive entropy and mutual information to help with the labelling process. We show how to bridge the gap between theory and practice by describing our pipeline from data preprocessing to model output visualisation. Additionally, we supply all of our data and code. The successful autonomous detection of mosquitoes allows us to perform analysis which is critical to the project goals of tackling mosquito-borne diseases such as malaria and dengue fever.

In view of the case that tolerance and maintenance design have been studied separately not to concentrate reflecting the relationship between process cost and product quality in the manufacturing systems, an optimization model of integrated tolerance and maintenance design for two-dimensional multi-station assembly processes is proposed. At first, with the relationship of dimension deviation propagation analyzed, a quality loss model for multi-station assembly processes is built; Then, a statistics feature model of deviation about fixture locating pin pairs is established, with the loss of the fixture process wear considered. At last the lifecycle cost of fixture system including fabrication and maintenance is defined, a two-dimensional multi-station assembly process optimization method was proposed, tolerance and maintenance parameters of the locating pin pairs in two-dimensional multi-station assembly process are optimized and designed. Automotive body side aperture assembly is given as an example to illuminate that the method is reasonable and effective.

The current work aims at determining optimal sensor configurations for the modal identification of a post-tensioned timber frame structure. The objective is to maximise the information gained from the structural testing, while keeping the number of necessary sensors to a minimum. Three different and widely used optimal sensor placement (OSP) methods are investigated and evaluated based on appropriate metrics, namely the effective independence method, the modal kinetic energy method and the information entropy (IE) method. An enhanced IE variant is adopted, which resolves the problem of close positioning of sensors, which forms a common issue when sensor positions are selected from a dense grid. In terms of the adopted metrics, three different options are investigated, namely, the information entropy index, the Modal Assurance Criterion and a newly introduced metric, the relative dispersion index. For the quantification of trade-offs among the selected metrics a Pareto Front scheme is realised. The study indicates that the evaluation of different OSP configurations is strongly dependent on the employed metric. It is therefore of vital importance to select the appropriate metric when determining optimal sensor positions.