Anomaly detection is one of the key applications of data utilization in smart factories, particularly in monitoring factory facilities. Early detection and resolution of anomalies, such as system failures, can lead to cost reduction and quality stabilization. One of the targets of abnormality detection applications in the industry section is a refrigerator unit used in food processing factories and warehouses. Anomalies in the early stages in refrigerator units appear in the operating sounds, which can enable their detection. In this study, we propose a method for detecting abnormal sound, extracting abnormal frequency components, and identifying the direction of the abnormal sound source. To identify the direction of the anomalous sound source, multi-channel sound recorded by a microphone array is used. To the best of our knowledge, no method has yet been proposed for anomaly sound detection using multi-channel acoustic data. In the proposed method, anomaly scores calculated in each channel of the microphone array are aggregated to determine whether the entire data is anomalous or not. Anomalous sounds were extracted from the anomaly data using a deep generative model. The extracted anomalous sounds were used to localize the sound source and the direction of the anomalous source was identified. The proposed method improved the precision of anomaly sound detection while maintaining the recall rate of a conservative comparison method. Using the proposed method, anomalous sounds were extracted from the anomaly data, and their arrival directions were identified.