Adaptive lift chiller units fault diagnosis model based on machine learning

Area of Science:

• Mechanical Engineering
• Artificial Intelligence

Background:

• Early-stage chiller faults are often imperceptible, leading to escalating issues.
• Traditional fault diagnosis methods lack accuracy and stability for detecting incipient faults.

Purpose of the Study:

• To develop an advanced fault diagnosis model for chillers.
• To enhance early fault detection capabilities in chiller units.

Main Methods:

• A novel Hybrid Improved Northern Goshawk Optimization Algorithm (HINGO) was developed, incorporating refraction opposition-based learning, sine-cosine strategy, Lévy flight, and a nonlinear decreasing factor.
• Least Squares Support Vector Machine (LSSVM) was optimized using the HINGO algorithm.
• An improved IAdaBoost ensemble learning algorithm was integrated with the HINGO-LSSVM model.
• The model was trained and validated using the ASHRAE RP-1043 air conditioning fault dataset.

Main Results:

• The HINGO-LSSVM-IAdaBoost model demonstrated superior performance in early fault diagnosis compared to traditional methods.
• The enhanced optimization strategies in HINGO improved population distribution and search capabilities.
• The ensemble approach provided robust and stable fault detection.

Conclusions:

• The proposed HINGO-LSSVM-IAdaBoost model offers significant advantages for early chiller fault diagnosis.
• This approach enhances the reliability and efficiency of chiller maintenance and operation.
• The study highlights the potential of hybrid optimization and ensemble learning in complex system diagnostics.