With the rapid advancement of digitalization and intelligence, robotics has garnered significant attention. The RV reducer, a core component in robotic joint applications and other high-precision transmission systems, has seen widespread adoption. However, its complex internal structure, demanding operational environments, and variable working conditions contribute to diverse failure modes. To prevent fault escalation and avoid economic losses, effective health monitoring and fault diagnosis are essential for ensuring reliable RV reducer operation. This article provides a comprehensive review of fault diagnosis methodologies for RV reducers. It details the approaches employed by researchers globally, categorizing them into methods based on dynamic model analysis, signal processing, and other techniques. The distinctions and applications of these different methods are systematically compared. Finally, the advantages and disadvantages of current RV reducer fault diagnosis methods are summarized, and prospective research directions are outlined.
The RV reducer is a closed, statically indeterminate, and compound planetary transmission mechanism. It typically consists of a first-stage involute planetary gear train and a second-stage cycloidal-pinwheel mechanism. The sun gear serves as the input, driving the planetary gears. The planetary gears are connected to crankshafts, which impart an eccentric motion to the cycloidal gears. The meshing between the cycloidal gears and the stationary pinwheel (housed in the pin housing) drives the planet carrier, which acts as the output. This intricate design, while providing high reduction ratios and compactness, also introduces multiple potential failure points, making fault diagnosis challenging.
1. Fault Diagnosis Based on Intrinsic Characteristics and Dynamic Modeling
Leveraging the inherent characteristics of the RV reducer, researchers construct dynamic models to analyze its kinetic behavior and operational patterns. Through modeling or simulation, the failure mechanisms of the RV reducer can be inferred. The general procedure for this analysis is as follows:
- Establish a dynamic model of the RV reducer.
- Simulate operational conditions, apply extreme stresses, and perform fatigue life analysis.
- Collect simulation signals from the dynamic model.
- Conduct fault diagnosis and analysis.
- Propose improvements for component performance.
Due to its low cost, dynamic analysis is commonly used for preliminary experimental assessment. The results can effectively predict component wear and are a standard method for preventive maintenance and fault diagnosis of the RV reducer. For instance, studies have developed equivalent tooth contact models to predict the fatigue life of the cycloidal-pinwheel pair under cyclic loading. Others have established system dynamics differential equations to calculate natural frequencies or created rigid-flexible coupled dynamic virtual prototype models to analyze vibration characteristics. Finite Element Analysis (FEA) is extensively used for modal analysis, harmonic response analysis, random vibration analysis, and thermal analysis of key components like the cycloidal gear and crankshafts, providing insights to avoid resonance, thermal deformation, and other failure modes.
2. Fault Diagnosis Based on Vibration Signal Analysis
During the operation of an RV reducer, its signal response comprehensively reflects its health status. Vibration signal analysis is the most prevalent method. However, the working conditions for an RV reducer are often complex and variable. Under constant conditions, Fourier Transform can identify fault-related energy peaks. Under variable speed conditions, signals become non-stationary, exhibiting frequency modulation, amplitude modulation, and phase modulation, which complicate analysis. This section categorizes and reviews vibration-based fault diagnosis methods for the RV reducer under different operational regimes.
2.1 Spectrum Analysis Methods
Spectrum analysis is a highly effective modern fault diagnosis technique. Vibration signals from an RV reducer are collected using a comprehensive test bench and analyzed. Since the fault characteristics of the reducer are constant, the fault characteristic frequencies of various components can be calculated based on the reducer’s geometry. By applying the Fast Fourier Transform (FFT) to the collected vibration signals and comparing the spectral peaks before and after a fault, the faulty component can be identified. For example, comparing the spectrum of a healthy RV reducer needle bearing with that of a faulty one reveals distinct energy peak changes at specific frequencies. Researchers have utilized this method with dedicated test rigs to understand the periodic characteristics of RV reducer vibration. Advanced approaches combine nonlinear output frequency response function (NOFRF) spectra with kernel principal component analysis (KPCA) to improve diagnostic accuracy over traditional spectral methods. Others analyze torsional vibration characteristics or use NOFRF spectra under harmonic excitation to build fault classifiers. Pre-processing techniques like the NeighCoeff method for filtering and denoising are also employed to enhance the clarity of fault information in the spectrum.
A critical limitation is that traditional FFT is primarily effective under steady-state, constant speed conditions. When the RV reducer operates under variable speed, the fault characteristic frequencies also change, leading to the “frequency smearing” phenomenon, which can cause misdiagnosis or missed faults if conventional spectrum analysis is applied directly.
2.2 Order Tracking Analysis Methods
Order tracking analysis is an effective method for monitoring and diagnosing faults in variable-speed machinery like the RV reducer. The core idea is to use signal processing algorithms to convert vibration signals sampled at equal time intervals into signals resampled at equal angular increments. Subsequent spectral analysis of this angle-domain signal yields an order spectrum, isolating speed-invariant fault signatures. The general flowchart involves signal acquisition, tachometer processing (or computed tachometer generation), angular resampling, and order spectrum calculation.
Order tracking can be hardware-based or computed. Hardware-based methods require additional tachometer hardware, which can be costly and difficult to install on compact RV reducers. Computed Order Tracking (COT) is therefore more prevalent, using numerical interpolation for angular resampling. COT itself is divided into methods with a tachometer (requiring careful sampling frequency selection based on speed change rate) and tacholess methods (which extract the fundamental frequency from the vibration signal itself based on transmission characteristics). For RV reducer diagnosis, tacholess order tracking combined with techniques like Empirical Mode Decomposition (EMD) or improved wavelet threshold denoising has been successfully applied to analyze non-stationary vibration signals from swing fatigue tests. By identifying characteristic orders for each component (e.g., planet gear order, bearing ball pass orders), fault locations can be pinpointed by comparing order magnitude changes. An improved computed order tracking method combined with angular domain synchronous averaging has also been proposed to effectively diagnose tooth root cracks in planet gears within the RV reducer.
2.3 Multi-Information Fusion and Intelligent Diagnosis Methods
The operational environment of the RV reducer is harsh, and its working conditions are complex. The multi-component, coupled transmission leads to monitoring information that is typically characterized by redundancy, multiple vibration sources, and significant noise interference. Therefore, multi-information fusion and intelligent diagnosis methods, which can process data comprehensively and accurately, have rapidly developed in this field. Algorithms with powerful data processing capabilities, such as deep learning, neural networks, and residual networks, have achieved significant success.
| Method Category | Specific Techniques / Models | Key Contribution / Application |
|---|---|---|
| Advanced Neural Networks | Improved Convolutional Capsule Network | Diagnosis of both single and compound faults in the RV reducer. |
| Hybrid Intelligent Models | EEMD with PSO-optimized Extreme Learning Machine (ELM) | Effective judgment of the RV reducer’s operational state. |
| Convolutional Neural Networks (CNN) | 2D-CNN, Noise-interference CNN, Denoising CNN | Improved diagnostic accuracy under noisy, variable speed, and variable load conditions. |
| Deep Feature Learning | Deep CNN combined with NOFRF spectra | Extraction and classification of hidden fault features from spectral images. |
| Residual Networks | ResNet-based models | Significantly enhanced fault diagnosis rates for the RV reducer. |
| Data Fusion Models | Multi-directional data fusion with Wavelet Convolution Energy Pooling Network | Effective fault diagnosis using single-point measurement data. |
| Other Intelligent Systems | SOM Neural Networks, ELM with Horizontal Learning Swarm Optimizer | Establishing fault recognition models and demonstrating good stability and generalization. |
These methods address the challenge of extracting weak fault features from low signal-to-noise ratio signals. However, issues like gradient vanishing, computational complexity, and the need for high accuracy remain. Therefore, fusing multiple models and algorithms to leverage their respective strengths is a promising direction for achieving more precise, reliable, and efficient diagnosis of the RV reducer.
3. Diagnosis Using Other Signal Modalities
3.1 Acoustic Emission (AE) Signal Analysis
Using the acoustic emission principle to detect internal faults in an RV reducer is a relatively new approach. The principle involves using AE sensors to detect transient elastic waves generated by the rapid release of energy from localized sources within the material, such as crack growth, wear, pitting, or breakage of components. This results in changes in contact stress, allowing for accurate fault localization. The monitoring and analysis methods are similar to vibration signal analysis, involving spectrum analysis, denoising, and signal reconstruction to locate the fault source. The propagation mechanism of AE signals within the reducer is complex, involving reflections and attenuations at interfaces. Researchers have explored methods based on Hidden Markov Models (HMM) for RV reducer fault detection using AE signals. Adaptive segmentation, frequency-domain transformation, and deep neural network analysis of AE signals, combined with angle data, have been used for fault point localization. Studies have also analyzed the共性 and差异 of AE signal characteristics for different basic fault types and investigated source localization techniques within bearings. Compressed sensing combined with wavelet decomposition and softmax classification has been proposed for RV reducer diagnosis and prediction using AE signals.
3.2 Motor Current Signature Analysis (MCSA)
Given that the RV reducer is a high-precision transmission device often driven by a servo motor, and considering the practical limitations of installing vibration sensors (space constraints, susceptibility to interference), some researchers utilize motor current signal analysis to locate faults in the RV reducer. The underlying principle is that mechanical faults in the driven load (the reducer) modulate the torque required from the motor, which in turn modulates the motor’s current. By analyzing these modulations, faults can be detected. Techniques include transforming current signals to the frequency domain and using optimized sparse autoencoders to extract fault features, applying discrete wavelet transform for time-frequency analysis and machine learning classification, and fusing current signals with vibration signals through methods like wavelet decomposition, Hilbert transform, and variational mode decomposition to obtain fault signatures. Experimental systems have also been developed to analyze the relationship between RV reducer transmission backlash and corresponding servo motor parameters (current, torque, speed, position) to diagnose fault types.
4. Special Considerations for RV Reducer Faults
The RV reducer, with its two-stage planetary-cycloidal design, has a more complex structure than traditional fixed-axis or simple planetary gear reducers. While traditional reducers commonly experience faults like gear tooth breakage or bearing race damage, the RV reducer has unique components prone to specific failures.
Crankshafts: As a critical link between the first and second stages, crankshafts are subject to bending, cracking, and breakage due to unbalanced loads, high friction, thermal stresses, and severe operating conditions (variable speed, high torque).
Needle Bearings: These are among the most susceptible components. Common failures include fatigue spalling due to cyclic contact stress, scuffing or adhesive wear from poor lubrication under high speed/load, and abrasive wear from contamination or vibration.
Cycloidal Gear: Failures include tooth breakage from excessive bending stress or progressive pitting, abrasive wear from friction debris, and scuffing from high meshing speeds and inadequate lubrication.
Output Needle Rollers: These rollers transmit force between the cycloidal disc and the output flange. They are prone to plastic deformation and wear due to uneven load distribution under eccentric motion, high loads, and thermal effects. Cage or separator failure can lead to roller misalignment and accelerated wear.
The complexity of the RV reducer’s internal dynamics and the low-frequency nature of some faults (e.g., cycloidal gear mesh frequency) make diagnosis challenging. Therefore, a combination of dynamic analysis for pre-design assessment and advanced signal processing or multi-information fusion for operational monitoring is often necessary.
5. Summary, Comparison, and Future Outlook
5.1 Summary and Comparative Analysis
In recent years, significant progress has been made in RV reducer fault diagnosis research. The table below summarizes the applicability, advantages, and disadvantages of the primary methods discussed.
| Fault Diagnosis Method | Key Advantages | Main Disadvantages / Challenges |
|---|---|---|
| Dynamic Model Analysis | Low cost, economically viable. Allows simulation of any component for pre-testing and failure mechanism insight. | Results may have errors due to model simplifications, assumptions about assembly precision, and differences from real operational conditions. |
| Spectrum Analysis | Method is straightforward. Only requires vibration signal measurement; fault location via peak comparison. | Primarily effective only for RV reducer under steady-state, constant speed conditions. |
| Order Tracking Analysis | Enables accurate diagnosis under variable speed conditions by converting to the angle domain. | High requirements for tachometer sampling frequency (if used) and computational algorithm robustness. Tacholess methods require accurate fundamental frequency extraction. |
| Multi-Information Fusion / Intelligent Analysis | Highly adaptable with many improvable techniques. Effectively processes complex data for comprehensive and accurate fault localization. | Algorithms can be complex, requiring significant computation time. May suffer from issues like gradient vanishing. Requires large, well-labeled datasets for training. |
| Acoustic Emission (AE) Analysis | Highly sensitive to incipient and microscopic faults (e.g., crack initiation). Non-intrusive; sensors mounted externally. | Sensors can be bulky. Signal propagation is complex (reflections, refraction, attenuation), making source localization difficult. Sensitive to external noise. |
| Motor Current Signature Analysis (MCSA) | Non-invasive; uses existing motor current sensors. Cost-effective for monitoring. | Signals are susceptible to interference from the motor itself and power supply. Fault signatures can be weak and less distinct for early-stage faults in the RV reducer. |
5.2 Future Research Directions
Despite the advancements, several areas require further investigation to enhance the reliability and practicality of RV reducer fault diagnosis.
1. Transmission Error Monitoring: Transmission error (TE), defined as the difference between the theoretical and actual output angular position, is a key indicator of RV reducer health and precision. Monitoring TE under operation and performing spectral analysis on the TE signal can reveal fault-related periodicities. The characteristic “orders” in the TE spectrum correspond to different components. By comparing the energy peaks of these orders before and after a fault, similar to order tracking, faults can be diagnosed. Future work should focus on high-precision, online TE measurement techniques and robust analysis methods for variable conditions.
2. Advanced Hardware-Assisted Order Analysis: Research into cost-effective, miniaturized hardware for direct angular sampling could revolutionize variable-speed diagnosis. Using a high-resolution rotary encoder (e.g., magnetic grating) pulse train as the trigger for vibration data acquisition enables true equal-angle sampling. The FFT of this directly acquired data yields an order spectrum without complex computational resampling, saving processing time and potential errors. The development and experimental validation of such integrated, low-cost hardware systems for the RV reducer is a promising direction.
3. Enhanced Robustness of Intelligent Fusion Methods: A critical challenge for data-driven methods is their robustness under strong noise and highly variable operating conditions. Excessive or insufficient denoising can distort signals, leading to模糊 or failed diagnosis. Future research should focus on developing intelligent models with inherent noise resistance and strong generalization capabilities across diverse and unseen operational profiles of the RV reducer. Furthermore, hybrid models that combine the strengths of physics-based models (e.g., dynamic models providing synthetic fault data) and data-driven models (for pattern recognition) could address the scarcity of real-world fault data and improve diagnostic accuracy and early warning capabilities.
4. Prognostics and Health Management (PHM) Integration: Moving beyond fault diagnosis to remaining useful life (RUL) prediction is the next frontier. Integrating the aforementioned diagnostic methods with degradation modeling and machine learning for prognostics will enable true predictive maintenance for the RV reducer, further reducing downtime and costs.
The pursuit of more accurate, robust, and practical fault diagnosis methods for the RV reducer remains essential to support the growing demands for reliability and longevity in robotics and advanced machinery.