قابلیت تشخیص و تجزیه و تحلیل حساسیت برای فرایندهای ماشینکاری چند ایستگاه

Dimensional variation is a major problem affecting product quality in discrete-part manufacturing. The stream of variation (SoV) methodology has been proposed as one of the systematic approaches to identify the root causes of process variation based on part measurements. This paper presents the results of the diagnosibility and sensitivity analysis study of the SoV methodology in a multi-station V8 cylinder head machining process used by a major domestic automotive manufacturer. The SoV model of dimensional machining errors has been derived based on the CAD description of the part and CAPP description of the process. Variation patterns of the final product were assessed based on the measurements of 20 automotive cylinder heads machined under normal process conditions, and the relative contributions of each machining station were assessed. In addition, one faulty product was observed and SoV model was used to identify the machining station that caused this quality problem. A station-level error decomposition method has been introduced and the SoV model correctly identified the culprit station. Furthermore, the sensitivities of dimensional features of the cylinder head to departures in fixture parameters away from their nominal values are evaluated based on the SoV model. Finally, four major issues arising from this implementation study of SoV in industry have been identified. Those are: non-diagnosibility of available measurements, vectorial representation of features, random sampling of parts at inspection and inadequate level of details in modeling of the process faults.

Dimensional quality problems due to the process variation are amongst the most critical issues for multi-station discrete part manufacturing, especially for precision parts such as engine blocks, heads or transmission components. Each manufacturing station in one such system introduces errors that propagate through the system and influence the final product quality. As indicated in Fig. 1, product quality errors x(k−1) accumulated in manufacturing stations 1, 2, 3,…, k−1 influence the product quality errors x(k) that are present after operations at manufacturing station k. In addition, at any manufacturing station k new errors u(k) are introduced which influence the outgoing product quality x(k). Measurements y(k) of the part quality can potentially be taken after operations at any station in order to depict the outgoing part quality. One should note that the inherent natural process variations W(k) also contribute to the product quality errors x(k) and thus also appear in the measured quality characteristics y(k). Reduction and elimination of the root-causes u(k) of quality problems that are introduced and accumulated in each manufacturing station k=1,2,…,N would lead to a reduction and elimination of the quality problems in the finished workpiece. Full-size image (16 K) Fig. 1. Dimensional errors flow in a multi-station manufacturing processes. Figure options Considerable efforts have been made to establish a mathematical connection between the root causes with the measured part quality characteristics [1], [2] and [3]. More recently, the so-called stream of variation (SoV) methodology [4] and [5] was introduced, which explicitly models the flow of dimensional errors from one station to another in the state-space form with the ordering index of the manufacturing station playing the role of the time index in the usual state-space models used in control theory. The reported research work in SoV methodology development includes four main aspects: (1) State-space modeling that links the product quality measurements with process faults [5], [6], [7] and [8]. (2) Algorithms for extracting fault information [9], [10] and [11]. (3) Design evaluation for product quality and process parameters [9] and [12]. (4) Evalutation and optimal selection of measurements and sensor locations [4], [13] and [14]. The aforementioned work pertaining to the SoV-based modeling and applications is mainly theoretical work and all the experimental results have been obtained from a controlled, laboratory environment. In [15], the authors reported a successful industrial implementation of the SoV idea in an automotive assembly plant. However, it utilized the knowledge-based product and process representation and pattern recognition rather than the explicit, analytical modeling of the process. There is currently no research that has been conducted on the application of the linear state-space model based SoV methodology in an uncontrolled, real factory environment. This paper reports the results of the diagnosibility and sensitivity analysis of the SoV methodology in a real industrial plant, where automotive cylinder heads are machined for one of the major domestic car manufacturers. It also discusses the issues raised from the application of the linear state-space SoV model in industrial environment. The remainder of this paper is organized as follows: Section 2 briefly reviews the SoV linear state-space modeling method for machining systems and describes the applications of this model in the industrial environment. Section 3 shows the SoV modeling results and the application results of this model in the V8 cylinder head machining process. The issues encountered during implementation of the SoV methodology in industry will be discussed in Section 4. Finally, Section 5 offers conclusions of the work presented in this paper and guidelines for possible future work.

This paper reports the results of the diagnosibility and sensitivity analysis study of the SoV methodology in a multi-station V8 cylinder head machining process. The SoV linear state-space model of dimensional machining errors based on the CAD description of the part and CAPP description of the process has been derived. Variation patterns of the final product were assessed based on measurements of 20 automotive cylinder heads machined under normal process conditions and within design specifications. Machining station that contributed to the deviation from nominal dimensions of one cylinder head has also been indicated using the SoV model-based station-level error decomposition method. In addition, sensitivities of dimensional features to departures in fixture parameters away from their nominal values have also been evaluated. Finally, four major issues associated with industrial implementation of the SoV methodology in machining have been identified and discussed. Those issues are lack of diagnosability of standard CAD product description and measurements encountered in contemporary machining systems, need for vectorial representation of workpiece features, random sampling of products for inspection at each inspection point in the system and a rather low level of detail in which SoV methodology currently models error sources in machining. In this paper, the non-diagnosability problem was circumvented with the use of the introduced station-level error decomposition method. Vectorial representations of workpiece features were obtained from manual transformation of the CAD description of the product and product measurements, while random sampling of products for inspection was overridden for 21 parts used in this study and their measurements were obtained from all the inspection stations used in this process. Finally, overcoming the weakness of the existing SoV models that are currently confined to dealing with only dimensional product characteristics and kinematic errors evaluated under the assumption of workpiece rigidity remains to be done through inclusion of non-dimensional part quality characteristics and workpiece compliance into the model. In addition, the generality and analytical tractability of SoV models allows for numerous applications of this methodology other than the ones presented in this paper. For example, the SoV model can be utilized to optimize measurements in multi-station machining in order to maximize the amount of information about the process-level root causes of dimensional machining errors [16] and [18]. Furthermore, sensitivity metrics can be employed for the optimal design of the process and fixtures, such that the sensitivity of the final product quality to errors in process parameters is minimized.