CONDITION MONITORING OF RECIPROCATING COMPRESSOR BASED ON VIBRATION CHARACTERISTIC USING ARTIFICIAL NEURAL NETWORKS

• Main Author: Bungaran, Nicolas
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/62677
• Institution: Institut Teknologi Bandung

Reciprocating compressor is one of the machines commonly used in various industries to produce compressed air by sucking and compressing the air with the reciprocating motion of the piston. One of the crucial activities in the operation of this reciprocating compressor is predictive maintenance using the condition monitoring method. This maintenance process uses input in the form of data taken from non-destructive measurement processes such as vibration measurements so that it does not cause machine downtime One of the disciplines that is developing in this era of computing is artificial intelligence methods. The artificial intelligence model that is quite commonly encountered today is the Artificial Neural Network (ANN) model. This method mimics the workings of the nervous system in the human brain to model the relationship between existing data and find certain patterns in the data. This Artificial Neural Network method can increase the reliability of diagnostic data as well as the speed of data diagnostics. One of the most used Artificial Neural Network algorithms is the back propagation model. The result of this undergraduate project is a backpropagation model Artificial Neural Network program that can predict damage to the piston compressor based on the symptoms found from the vibration measurement process. The test shows that the Artificial Neural Network program trained with Bayesian Regularization and with 22 iterations succeeded in predicting 200 tested vibration data with accuracy of 95.31%.