Signature construction and matching for fault diagnosis in manufacturing processes through fault space analysis

Variation-source identification in manufacturing processes is highly desired since it enables improvements in product quality. Recently, data-driven variation-source identification has received considerable attention. This paper presents a systematic variation-source identification method by assuming a linear model between the quality measurements and process faults. The noise term in the model is assumed to have a simple form. The variation-source identification is achieved through the testing of the common eigenspace between the fault signatures and the covariance matrix of the newly collected samples. Three types of fault signatures are constructed from either one or two covariance matrices for pattern matching. A systematic procedure to construct the signature is presented. A case study of a machining operation is conducted to illustrate the effectiveness of the proposed methodology.