What question did this study set out to answer?

The aim is to improve fault diagnosis accuracy of diesel engine motors by employing advanced signal processing and machine learning techniques.

April 8, 2026Discover Applied Sciences0 citationsOpen Access

Fault diagnosis of diesel engine motor systems based on empirical pattern decomposition and support vector machines

UUtsavMaharshi Dayanand University RKRaj KumarMaharshi Dayanand University ADArvind DhakaManipal Academy of Higher Education

Key Points

The aim is to improve fault diagnosis accuracy of diesel engine motors by employing advanced signal processing and machine learning techniques.
Implemented a hybrid CEEMD–TQWT framework for vibration signal analysis.
Utilized statistical energy and entropy descriptors from decomposed signals as inputs.
Employed Multi-Kernel Support Vector Machine (MKSVM) for classification of fault types.
Achieved a fault diagnosis accuracy improvement from 77.5% to 87.6%.
Demonstrated high reliability with a cross-validation variance of less than 1.8%.
Outperformed single-kernel SVMs (accuracy ≤ 76.5%) and EMD methods (accuracy ≈ 73%).

Abstract

Motor systems in Diesel engines exhibit complex vibration characteristics due to coupled mechanical–electrical interactions. Faults such as bearing degradation, bolt looseness, and shaft imbalance cause non-stationary vibration components across multiple frequency bands. The conventional method of spectrum analysis struggles to isolate the unique component of the problem in motor fault diagnosis. The current fault diagnostic approaches are inadequate in distinguishing among the many failure modes associated with the fault component. This paper proposes a hybrid CEEMD–TQWT–MKSVM framework for precise fault diagnosis of Diesel engine motors. Complementary Ensemble Empirical Mode Decomposition (CEEMD) suppresses mode mixing and decomposes the raw vibration signals into seven intrinsic mode functions (IMFs), while the Tunable Q-Factor Wavelet Transform (TQWT) adaptively enhances transient impact and harmonic features. Statistical energy and entropy-based descriptors are extracted from each subband and used as input to a Multi-Kernel Support Vector Machine (MKSVM) classifier combining RBF and Polynomial kernels. The results of the experiments demonstrate that the combination of CEEMD and TQWT is capable of efficiently removing modal aliasing and obtaining IMF components that clearly characterize the fault characteristics. When compared to the approaches that are currently in use, the Multi-kernel support vector machine has shown an improvement in fault diagnostic accuracy, which has increased from 77.5% to 87.6%. Experimental analysis over 34 signal groups under four operating conditions (normal, imbalance, bolt loosening, bearing wear) shows that the proposed model achieves an average classification accuracy of 87.6%, outperforming single-kernel SVMs (accuracy ≤ 76.5%) and EMD-based approaches (accuracy ≈ 73%). The method also demonstrates a cross-validation variance < 1.8%, indicating high reliability. The results confirm that CEEMD–TQWT enhances feature separability in time–frequency space, while MKSVM provides stable fault recognition suitable for real-time Diesel engine monitoring.

问 AI

Bookmark

View Full Paper