Supply chain management, as one of the most important aspects of business management, plays a key role in the effective communication between manufacturers, suppliers, and customers. With the rapid advancements in communication and computer technologies, more precise and efficient supply chain management has become possible something that was not achievable before the 1980s. Today, organizations can leverage modern techniques and technologies to create highly competitive supply chains, enabling them to quickly respond to market demand changes and perform more successfully in highly competitive markets. One of the fundamental challenges in supply chain management is the bullwhip effect, where demand fluctuations amplify as they move up the chain. This phenomenon can lead to excessive increases or decreases in inventory, storage costs, uncertain production planning, and ultimately reduced supply chain efficiency. Defects in spare parts can exacerbate the bullwhip effect by creating unexpected demand fluctuations, which are difficult to predict and manage. In fact, when a spare part in a rotating machine fails, in the absence of prior planning, production and distribution delays occur, ultimately leading to an increase in emergency orders and a lack of coordination between different units. In this research, a hybrid Convolutional Neural Network and Long Short-Term Memory (CNN-LSTM) network is employed to predict and identify defects in spare parts of rotating machines using signal data, as well as to predict demand fluctuations. The results of the proposed model are then compared with machine learning algorithms, including RF, SVM, KNN, and MLP, based on eva‎luation metrics. Since the data used in this study are signal-based, features need to be extracted from these signals for use in machine learning models, for which the Fourier transform is employed. The eva‎luation results of different models show that the hybrid model, with an accuracy of 0.93 in F1_Score and 0.93 in Precision, performs better compared to the RF model with 0.79 in F1_Score and 0.79 in Precision, the SVM model with 0.82 in F1_Score and 0.79 in Precision, the KNN model with 0.78 in F1_Score and 0.78 in Precision, and the MLP model with 0.89 in F1_Score and 0.89 in Precision.

ناصر ملاوردي اصفهاني

صبا صارمي نيا , مهدي ايران پور