| Traditional hard gauge checking fixtures or Coordinate Measuring Machines (CMM) cannot provide large enough samples for effective Statistical Process Control (SPC) in automobile body assembly due to their off-line nature and low speed. With in-line Optical Coordinate Measuring Machines (OCMM), every body assembled can be measured, resulting in 100% measurement. However, manufacturers fail to make efficient use of the data.;Conventional control charts, e.g., ;Process monitoring. Autocorrelation in data can result in false alarms when control charts are directly applied to data. The application of Prediction Error Analysis (PEA) can reduce the false alarm rate and also affect the detection speed. The effect of PEA on detection speed is analyzed and presented with examples based on AR(1) and ARMA(2,1) models for a step-function type mean shift.;Process parameter identification. Sources of dimensional variation can be identified from the 100% measurement data. Using the autocorrelation in data, process physical characteristics, such as natural frequency, can be estimated. The contribution of each dynamic mode to the total variation can be quantitatively analyzed through decomposition of autocovariance. Cross-correlation can be used to reveal inter-sensor relationships or deformation patterns, such as Side Frame Misalignment or "Match-Boxing".;Process variation reduction. "Adaptive quality control" using Forecasting Compensatory Control (FCC) is presented using simulation. However, due to lack of control mechanisms that actuate control instantly, body assembly process can only be adjusted on a batch-to-batch basis. Process control is based on the detection of process faults and human interference. Two successful case studies in variation reduction are presented. |
