Open Access

ARTICLE

Acoustic Noise-Based Scroll Compressor Diagnosis during the Manufacturing Process

Geunil Lee¹, Daeil Kwon^2,*

1 Department of Industrial Engineering, Sungkyunkwan University, Suwon, 16419, Republic of Korea
2 Department of Systems Management Engineering, Sungkyunkwan University, Suwon, 16419, Republic of Korea

* Corresponding Author: Daeil Kwon. Email:

(This article belongs to the Special Issue: Data-Driven and Physics-Informed Machine Learning for Digital Twin, Surrogate Modeling, and Model Discovery, with An Emphasis on Industrial Applications)

Computer Modeling in Engineering & Sciences 2025, 144(3), 3329-3342. https://doi.org/10.32604/cmes.2025.069402

Received 22 June 2025; Accepted 04 September 2025; Issue published 30 September 2025

Abstract

Nondestructive testing (NDT) methods such as visual inspection and ultrasonic testing are widely applied in manufacturing quality control, but they remain limited in their ability to detect defect characteristics. Visual inspection depends strongly on operator experience, while ultrasonic testing requires physical contact and stable coupling conditions that are difficult to maintain in production lines. These constraints become more pronounced when defect-related information is scarce or when background noise interferes with signal acquisition in manufacturing processes. This study presents a non-contact acoustic method for diagnosing defects in scroll compressors during the manufacturing process. The diagnostic approach leverages Mel-frequency cepstral coefficients (MFCC), and short-time Fourier transform (STFT) parameters to capture the rotational frequency and harmonic characteristics of the scroll compressor. These parameters enable the extraction of defect-related features even in the presence of background noise. A convolutional neural network (CNN) model was constructed using MFCCs and spectrograms as image inputs. The proposed method was validated using acoustic data collected from compressors operated at a fixed rotational speed under real manufacturing process. The method identified normal operation and three defect types. These results demonstrate the applicability of this method in noise-prone manufacturing environments and suggest its potential for improving product quality, manufacturing reliability and productivity.

---

Keywords

---