Multistage Manufacturing Processes Research Articles

Monitoring and diagnosing of mean shifts in multivariate manufacturing processes using two-level selective ensemble of learning vector quantization neural networks

Nowadays, modern computers in general and the PC in particular have made the continuous high-speed acquisition and inspection accessible and during the last decade, multivariate control charts were given more attention and became so popular in real-world manufacturing scenarios. However, most conventional multivariate control charts share the same problem: they can only detect an out-of-control circumstance but cannot directly point out which variable or group of variables has caused the out-of-control signal. This study proposes an effective MSPC model enabled by two-level discrete particle swarm optimization-based selective ensemble of learning vector quantization networks (DPSOSENLVQ) for monitoring and diagnosing of mean shifts in multivariate manufacturing processes. In this model, one DPSOSENLVQ is developed for detecting out-of-control signals in process mean, while the other DPSOSENLVQ is developed for further classifying the detected out-of-control signals as one of the specific mean shift types. The experimental result indicates that the proposed MSPC model can not only efficiently monitor the process state but also accurately diagnose the detected out-of-control signals. Empirical comparisons also showed that the proposed MSPC model outperformed other existing approaches in literature. In addition, a case study is conducted to demonstrate how the proposed MSPC model can function as an effective tool for monitoring and diagnosing of mean shifts in multivariate manufacturing processes.

Diagnosing Multistage Manufacturing Processes With Engineering-Driven Factor Analysis Considering Sampling Uncertainty

A new engineering-driven factor analysis (EDFA) method has been developed to assist the variation source identification for multistage manufacturing processes (MMPs). The proposed method investigated how to fully utilize qualitative engineering knowledge of the spatial variation patterns to guide the factor rotation. It is shown that ideal identification can be achieved by matching the rotated factor loading vectors with the qualitative indicator vectors (IV) that are defined according to spatial variation patterns based on the design constraints. However, the random sampling variability may significantly affect the estimation of the rotated factor loading vectors, leading to the deviations from their true values. These deviations may change the matching results and cause misidentification of the actual variation sources. By using implicit differentiation approach, this paper derives the asymptotic distribution and the associated variance-covariance matrix of the rotated factor loading vectors. Therefore, by considering the effect of sample estimation variability, the variation sources identification problem is reformulated as an asymptotic statistical test of the hypothesized match between the rotated factor loading vectors and the indicator vectors. A real-world case study is provided to demonstrate the effectiveness of the proposed matching method and its robustness to the sample uncertainty.

Multimode variation modeling and process monitoring for serial-parallel multistage manufacturing processes

A Serial-Parallel Multistage Manufacturing Process (SP-MMP) may have multiple variation propagation modes in its production runs when process routes vary from part to part. Conventional methods that ignore such multimode variation may not be able to effectively model and monitor the variation streams. It is also very challenging to model such a process when the available engineering domain knowledge is insufficient to characterize the variation streams. This article proposes a data-driven method, piecewise linear regression trees, to interrelate the variables for an SP-MMP with multimode variation. A unified control chart system is developed to monitor the process considering modeling uncertainty. The application to a more generic multistage multimode process is discussed. Finally, the effectiveness of the proposed procedure is demonstrated in an application involving a wafer manufacturing process.

Sample Size Determination in NT-Net Quasi-Monte Carlo Simulation

Monte Carlo (MC) technique prevails in probabilistic design simulation, such as in statistical tolerance analysis and synthesis. A quasi-Monte Carlo (e.g., number theoretic net method (NT-net)) with better computation efficiency over MC has recently attracted interests in application. In spite of a comprehensive case study (Huang et al., 2004, “Tolerance Analysis for Design of Multistage Manufacturing Processes Using Number-Theoretical Net Method (NT-net),” Int. J. Flexible Manuf. Syst., 16, pp. 65–90 and Zhou et al., 2001, “Sequential Algorithm Based on Number Theoretic Method for Tolerance Analysis and Synthesis,” ASME J. Manuf. Sci. Eng., 123(3), pp. 490–493) for comparison between NT-net and MC, a method for sample size determination of NT-net is still not available. Combinatorial theory and the solution of occupancy problem are used for estimating equivalent sample sizes of MC and NT-net, allowing the NT-net sample size determination in application. A multivariate Chebyshev polynomial with variant coefficients is used to represent generic design functions for validation. The results are verified by case studies.

Process-oriented tolerancing using the extended stream of variation model

Current works on process-oriented tolerancing for multi-station manufacturing processes (MMPs) have been mainly focused on allocating fixture tolerances to ensure part quality specifications at a minimum manufacturing cost. Some works have also included fixture maintenance policies into the tolerance allocation problem since they are related to both manufacturing cost and final part quality. However, there is a lack of incorporation of other factors that lead to increase of manufacturing cost and degrade of product quality, such as cutting-tool wear and machine-tool thermal state. The allocation of the admissible values of these process variables may be critical due to their impact on cutting-tool replacement and quality loss costs. In this paper, the process-oriented tolerancing is expanded based on the recently developed extended stream of variation (SoV) model which explicitly represents the influence of machining process variables in the variation propagation along MMPs. In addition, the probability distribution functions (pdf) for some machining process variables are analyzed, and a procedure to derive part quality constraints according to GD&T specifications is also shown. With this modeling capability extension, a complete process-oriented tolerancing can be conducted, reaching a real minimum manufacturing cost. In order to demonstrate the advantage of the proposed methodology over a conventional method, a case study is analyzed in detail.

Method of Identifying Key Quality Characteristics in Multistage Manufacturing Process Based on Variation Transmission Model

Method of Identifying Key Quality Characteristics in Multistage Manufacturing Process Based on Variation Transmission Model

Reduction of Springback of High-strength Car Body Structural Components

High-strength car body structural components are used in vehicles to meet the increasing crash resistance and lightweight construction requirements. In most cases, they are produced in a multi-stage manufacturing process. The problem with this trend is that the tendency to spring back increases as the materials are becoming stronger. The measures taken so far to optimise dimensional stability involve some disadvantages. Therefore, an innovative method of reducing springback is to be de veloped at the Ostfalia University of Applied Sciences. The pre-drawing stage is modified in the forming simulation to purposefully influence the state of stress and thus elastic recovery.

On-line classifying process mean shifts in multivariate control charts based on multiclass support vector machines

In multivariate statistical process control (MSPC), most multivariate control charts can effectively monitor anomalies based on overall statistic, however, they cannot provide guidelines to classify the source(s) of out-of-control signals. Classifying the source(s) of process mean shifts is critical for quality control in multivariate manufacturing process since the immediate identification of them can greatly help quality engineer to narrow down the set of possible root causes and take corrective actions. This study presents an improved particle swarm optimisation with simulated annealing-based selective multiclass support vector machines ensemble (PS-SVME) approach, in which some selective multiclass SVMs are jointly used for classifying the source(s) of process mean shifts in multivariate control charts. The performance of the proposed PS-SVME approach is evaluated by computing its classification accuracy. Simulation experiments are conducted and a real application is illustrated to validate the effectiveness of the developed approach. The analysis results indicate that the developed PS-SVME approach can perform effectively for classifying the source(s) of process mean shifts.

Method of Identifying Key Quality Characteristics in Multistage Manufacturing Process Based on PLSR

To improve the existing methods of identifying the key quality characteristics in multistage manufacturing process, the partial least squares regression (PLSR) method is combined with the state space model that a new method of identifying the key quality characteristics in multistage manufacturing process based on PLSR is proposed. According to the feature of multistage manufacturing process, the state space model is introduced to build the key quality characteristics identifying model for multistage manufacturing process, using the PLSR method to solve the problem of the quality characteristics such as multicollinearity, do model analyzing and identify the key quality characteristics. At last, the cigarette production process is presented as an example to introduce the application of this method. The result shows that this method can not only identify the key quality characteristics in multistage manufacturing process, but also establish the model of output quality effecting of all levels on the final product quality and its quality characteristics relationship, which reflect the structure of the multistage manufacturing process and causal relationship between quality characteristics at all process levels, provide the basis for quality analysis and control in multistage manufacturing process.

Using frailty models to account for heterogeneity in multistage manufacturing and service processes

Statistical process control has been widely applied to manufacturing and service operations with the aim of monitoring and improving the reliability of products. The existing monitoring procedures were introduced following the assumption that a single-stage process with independent quality characteristic is under consideration. However, in multistage processes with dependent variables, quality characteristics of interest should be optimally monitored only after they have been adjusted for the effect of influential covariates. In general, it is impossible to include all relevant covariates because measuring such values entails great financial costs. The neglect of such covariates results in having unobserved heterogeneity which dampens the detection ability of the monitoring procedure. The more complicated picture arises when the values corresponding to the reliability-related quality variable are censored due to the time and cost constraints. Thus, to deal with the effect of observed and unobserved covariates together with the censoring issue, the frailty and the proportional hazard models are used and some model-based monitoring schemes are devised. The surveillance procedures are proposed in both the presence and absence of a censoring mechanism. The performance analysis shows that the monitoring procedure based on the cumulative sum chart is superior in detecting shifts while the exponentially weighted moving average chart is effective in some cases.

Sample size determination for estimating multivariate process capability indices based on lower confidence limits

With the advent of modern technology, manufacturing processes have become very sophisticated; a single quality characteristic can no longer reflect a product's quality. In order to establish performance measures for evaluating the capability of a multivariate manufacturing process, several new multivariate capability (NMC) indices, such as NMC p and NMC pm , have been developed over the past few years. However, the sample size determination for multivariate process capability indices has not been thoroughly considered in previous studies. Generally, the larger the sample size, the more accurate an estimation will be. However, too large a sample size may result in excessive costs. Hence, the trade-off between sample size and precision in estimation is a critical issue. In this paper, the lower confidence limits of NMC p and NMC pm indices are used to determine the appropriate sample size. Moreover, a procedure for conducting the multivariate process capability study is provided. Finally, two numerical examples are given to demonstrate that the proper determination of sample size for multivariate process indices can achieve a good balance between sampling costs and estimation precision.

Reconfigured piecewise linear regression tree for multistage manufacturing process control

In a multistage manufacturing process, extensive amounts of observational data are obtained by the measurement of product quality features, process variables, and material properties. These data have temporal and spatial relationships and may have a non-linear data structure. It is a challenging task to model the variation and its propagation using these data and then use the model for feedforward control purposes. This article proposes a methodology for feedforward control that is based on a piecewise linear model. An engineering-driven reconfiguration method for piecewise linear regression trees is proposed. The model complexity is further reduced by merging the leaf nodes with the constraint of the control accuracy requirement. A case study on a multistage wafer manufacturing process is conducted to illustrate the procedure and effectiveness of the proposed method.

An Adaptive Multivariate Control Chart for Individual Observations with Adaptive EWMS Star Glyphs for Diagnosis

A time-weighted multivariate control chart for individual observations is introduced to monitor process variability changes in multiple quality characteristics in phase II applications. Because of adaptive adjustments, the proposed control chart is sensitive to total variation increase in a multivariate manufacturing process. A new visualization tool, named adaptive EWMS star glyphs, is proposed for process diagnosis once an out-of-control signal is found. An example is presented to illustrate the application of the proposed charting procedure.

Improving the Product Reliability in Multistage Manufacturing and Service Operations

Monitoring and improving the product reliability is of main concern in a large number of multistage manufacturing processes. The process output is commonly inspected under limited load conditions, and the tensile strength of reliability‐related quality characteristic is measured. This brings about censored observations that make the direct application of traditional control charts futile. The monitoring procedure becomes aggravated when the influence of variable competing risk is pronounced during the conducted test. To deal with this critical issue, we propose a regression‐adjusted cumulative sum (CUSUM) chart to effectively monitor a quality characteristic that may be right censored because of both fixed and variable competing risks. Moreover, two exponentially weighted moving average (EWMA) control charts on the basis of conditional expected values are devised to detect decreases in the tensile mean. The comparison of the three competing monitoring schemes confirms the superiority of the regression‐adjusted CUSUM procedure. Not only is the proposed control chart applicable to manufacturing processes with the aim of monitoring reliability‐related quality variables, it is also appropriate for monitoring similar quality measurements in service operations such as survivability measures in healthcare services. Copyright © 2011 John Wiley & Sons, Ltd.

Chart allocation strategy for serial–parallel multistage manufacturing processes with multiple faults

Although statistical process control (SPC) charts have been used for multistage manufacturing process quality monitoring for years, how to effectively allocate conventional SPC charts appropriately in a serial–parallel multistage processes has not been thoroughly studied. In this article, aiming to minimize the expected overall quality cost, we use the linear state-space model to model multistage processes and propose a strategy to properly allocate control charts in serial–parallel multistage manufacturing processes by considering the interrelationship information between stages and the simultaneous occurrences of multiple faults at different workstations. Based on the proposed chart allocation strategy, we are able to make rational chart allocation decisions to achieve quicker detection capability over the whole potential fault set. A hood assembly example is used to demonstrate the applications of the chart allocation strategy.

Compensability of errors in product quality in multistage manufacturing processes

Recent advances in model-based feedforward control of quality in multistage manufacturing processes (MMPs) are used to introduce the concept of compensability of errors in product quality. It is defined as the system-level ability to automatically eliminate quality errors accumulated in the product, given the allocation of measurements and controllable tooling. Significant similarities are observed with the control theory concepts of controllability and observability. In addition, just like in control theory, it is shown that relatively simple matrix rank tests can be used to determine the compensability of a MMP and are related with the previously introduced notion of diagnosability. The theory introduced in this paper enables rapid quantitative analysis of one's ability to actively control the quality in a manufacturing system, which is crucial for optimal allocation of measurements and controllable tooling in an MMP. Compensability studies based on the newly introduced methods are conducted using models of the flow of errors in product quality in two automotive cylinder head machining lines and a multi-layer semiconductor lithography process.

Stochastic Control of Multilayer Overlay in Lithography Processes

The multilayer overlay lithography process is one of the most important steps in semiconductor manufacturing. It is a multistage manufacturing process in which errors are introduced in each step, and existing errors accumulated in upstream operations cause occurrence of additional errors in downstream operations. Thus, variations in the previously deposited reference layers will also affect the overlay errors in the subsequently deposited layers. Therefore, a systematic method is needed to model the flow of overlay errors from layer to layer and to mitigate this effect using controllable process parameters. In this paper, a stochastic model for the introduction, accumulation and propagation of multilayer overlay errors of the lithography process is proposed. Based on that model, a sequential optimization method is utilized to find the optimal adjustments of controllable process parameters for minimizing the total multilayer overlay errors. A number of simulation case studies are presented to demonstrate the newly introduced model-based approach to controlling the multilayer overlay errors in lithography processes.

Design of multi-station manufacturing processes by integrating the stream-of-variation model and shop-floor data

Abstract Process design has been intensively studied to reduce dimensional variability of products produced in multi-station manufacturing processes (MMPs). Most of the existing studies focus on predicting variation propagation and evaluating process robustness. However, these studies overlook the potential use of historical shop-floor quality data of existing MMPs in order to extract the actual manufacturing operation capabilities from each station, and then, to evaluate more accurately the expected dimensional variability of new candidate process plans. This paper proposes a methodology to improve process plan selection based on three components: (i) based on historical shop floor data, an inference on the process capabilities of the stations in an existing MMP, which will be used to produce the new product; (ii) a sensitivity analysis of candidate process plans to identify critical fixtures and manufacturing stations/operations; and (iii) an optimal selection of candidate process plans. A case study is presented to demonstrate the effectiveness of the methodology.

On-line analysis of out-of-control signals in multivariate manufacturing processes using a hybrid learning-based model

Advanced automatic data acquisition is now widely adopted in manufacturing industries and it is common to monitor several correlated quality variables simultaneously. Most of multivariate quality control charts are effective in detecting out-of-control signals based upon an overall statistics in multivariate manufacturing processes. The main problem of such charts is that they can detect an out-of-control event but do not directly determine which variable or group of variables has caused the out-of-control signal and what is the magnitude of out of control. This study presents a hybrid learning-based model for on-line analysis of out-of-control signals in multivariate manufacturing processes. This model consists of two modules. In the first module using a support vector machine-classifier, type of unnatural pattern can be recognized. Then by using three neural networks for shift mean, trend and cycle it can be recognized magnitude of mean shift, slope of trend and cycle amplitude for each variable simultaneously in the second module. The performance of the proposed approach has been evaluated using two examples. The output generated by trained hybrid model is strongly correlated with the corresponding actual target value for each quality characteristic. The main contributions of this work are recognizing the type of unnatural pattern and classification major parameters for shift, trend and cycle and for each variable simultaneously by proposed hybrid model.

Functional Morphing in Multistage Manufacturing and Its Applications in High-Definition Metrology-Based Process Control

In multistage manufacturing processes equipped with high-definition metrology (HDM), part surface quality characteristics can be observed to change or “morph” from stage to stage. Such part surface variation propagations are caused by the physical processes, part attributes, and the interaction between stages. Previous research on variation propagation modeling focuses on part dimensional quality using discrete key product characteristics or vectors which have limitations in analyzing complex surface variation patterns contained in the HDM data. This paper proposes a new concept of functional morphing to characterize the surface changes and applies it to process control in high-precision manufacturing. Unlike conventional morphing algorithms that focus on transformations between geometries only, functional morphing integrates process physical insights into the geometric mappings, thus characterizing the complex HDM data patterns in physically meaningful ways. Specifically, a functional free form deformation approach including forward and backward mappings is developed to extract mapping functions between manufacturing stages to enable surface variation propagation analysis. The forward mapping function allows for accurate interstage adjustment that introduces shape deformation upstream to compensate for the end-of-line errors. The backward mapping function can predict surfaces at intermediate stages based on end-of-line measurements, leading to a cost-effective interstage process monitoring scheme. The interstage monitoring can also ensure the repeatability of a process controlled by the interstage compensation algorithm. The developed monitoring and adjustment methods are demonstrated via a case study of a two-stage machining process. Other potential applications of functional morphing such as process tolerance design are also discussed.