Abstract: Compressor is the core component of refrigeration equipment and the main source of equipment noise. The abnormal detection of product quality based on big data is a typical demand of industrial intelligence. It is based on industrial data and artificial intelligence methods, and realizes the abnormal detection of products by establishing data-driven models. High quality labeled samples are the key to compressor anomaly detection using artificial intelligence methods. An autoencoder-based automatic sample labeling method is proposed. Based on the sample reconstruction error sequence of the autoencoder, the samples are gradually adjusted, and the training samples are gradually extracted. Finally, normal samples and fault samples with accurate labeling are extracted from a large number of samples with labeling errors. This method solves the problem of real-time labeling of large-scale samples manually, and can provide accurate labeling training samples for online anomaly detection of products.

Key words: Sample labeling, Auto Encoder, Progressive adjustment, Unsupervised, Compressor