Abstract: Looseness faults are widely present in mechanical equipment, and diagnosing these faults under variable speed conditions remains challenging. To address this problem, we propose a time–frequency modal decomposition method optimized for variable speeds. To enhance the adaptability of this method across different operating conditions, we have optimized the parameter window width, obtaining an adaptive time–frequency modal decomposition technique. We conducted experimental validations on both bearing seat looseness faults and rotating component looseness faults. In the bearing seat looseness fault experiment, we analyzed time–domain waveforms, time–frequency plots, and the outputs of the adaptive time–frequency modal decomposition. This comprehensive analysis allowed us to accurately diagnose the looseness fault. In the rotating component looseness fault experiment, we processed variable speed bearing vibration signals. The proposed method effectively extracted fault features from these signals. Experimental results confirmed the effectiveness of our proposed method. Both bearing seat looseness and rotating part looseness faults were diagnosed using the adaptive time–frequency modal decomposition algorithm. Our experiments demonstrated that the proposed method can achieve modal decomposition of nonstationary signals, showcasing its effective diagnostic capabilities for looseness faults. Overall, this research highlights the effectiveness of the adaptive time–frequency modal decomposition method in diagnosing looseness faults under variable speeds and strong noise backgrounds. The method is applicable to different types of looseness faults and exhibits adaptability by optimizing parameters to cope with different noise intensities. It is important to note that this method requires instantaneous speed information from the device, which may impose some limitations on its applicability. However, in practical engineering, the technical challenge of measuring speed is relatively low. Many types of existing large-scale equipment are already equipped with speed measurement devices, providing a hardware foundation for implementing this method. Therefore, the aforementioned technical limitations can be easily resolved. Theoretical analysis and experimental validation in this study indicate that this innovative approach expands the application potential in the field of signal decomposition, providing a powerful tool for the health monitoring of mechanical equipment.