Signal feature extraction for sewing analysis using non-linear techniques

This paper describes the measurement of a process variable in industrial sewing machines using a piezoelectric sensing system and non-linear filtering and processing techniques. The objective is to obtain a measure of the needle penetration and withdrawal forces in a fabric, which can be used to detect faulty processing conditions causing second-rate quality products. Aimed for future real-time applications, currently the measurement process has been optimized for laboratory use. It provides the means for experimentation leading to the development of further processing techniques able to detect fault conditions automatically, a completely innovative industrial application. The signal is acquired by a piezoelectric sensor introduced in the needle support bar. With this set-up, the sensor provides not only the forces related to needle-fabric interaction, but also other, undesired components, that are reduced with specifically developed techniques. This paper presents the advantages and drawbacks of the measurement set-up and describes the optimization of the complex processing techniques used to extract the relevant information from the signals.