In the field of manufacturing, the widespread application of industrial robots has become a core driver for enhancing production efficiency, optimizing cost structures, ensuring product quality, and improving operational safety. Their continuous advancement in intelligence and automation is leading a profound transformation in modern industrial production models. However, once an industrial robot system fails, it can directly cause production line interruptions, delaying production cycles, increasing defect rates, and resulting in significant economic losses. The RV reducer, as a key component of industrial robots, is widely used in heavy-duty areas such as robot shoulders due to its excellent load-bearing capacity and high rigidity. The performance of this component not only directly determines the positioning accuracy and repeatability of industrial robots but also deeply affects the overall production efficiency and final product quality of equipment manufacturing enterprises. The optimization of its technical parameters is of irreplaceable strategic value for ensuring the stable operation of modern manufacturing. With the rapid development of industrial internet and artificial intelligence technologies, building real-time and intelligent fault diagnosis data acquisition platforms has become an important breakthrough in improving equipment health management.
Traditional fault diagnosis data acquisition systems often rely on single-type sensor data, and platform designs lack compatibility, typically targeting only specific types of reducers. In this work, we design and develop an integrated multi-sensor data acquisition platform to dynamically capture multi-dimensional operational data such as vibration, temperature, and noise from the RV reducer. This provides a high-quality data foundation for establishing deep learning-based fault prediction models, which holds significant engineering value for promoting the transition of industrial robot maintenance modes to predictive maintenance, reducing the lifecycle maintenance costs of key components, and ensuring the continuous and efficient operation of smart manufacturing production lines.
The RV reducer exhibits high complexity due to its multi-level composite structure, with core components consisting of precision mechanical parts working in synergy. The main modules include the central gear shaft, planetary gear set, double cycloidal pinwheel, pin gear pair, and power output shaft. To systematically understand the RV reducer, we summarize its basic structure and common failure modes, which are critical for designing an effective data acquisition platform.
The common failures of the RV reducer in industrial environments are primarily caused by long-term exposure to alternating loads, impact vibrations, assembly stresses, and other factors, leading to gradual performance degradation and functional failure of key components. Based on our analysis, we categorize typical faults into several types, as shown in the table below.
| Fault Type | Description | Typical Symptoms |
|---|---|---|
| Planetary Gear System Fault | Fatigue pitting, partial spalling, adhesive wear, and tooth breakage due to cyclic alternating loads. | Increased noise, reduced transmission accuracy stability. |
| Cycloidal Gear System Fault | Fatigue fracture, tooth surface gluing, contact fatigue-induced pitting, tooth surface wear, and plastic deformation. | Abnormal vibrations, efficiency drop. |
| Crankshaft Fault | Fatigue bending deformation, structural instability, and crankshaft fracture related to eccentric loads. | Misalignment, severe vibrations. |
| Bearing Fault | Wear or fracture of rollers or balls, cage fracture due to axial and radial forces. | Increased friction, overheating. |
| Other Faults | Seal wear and aging, bolt preload insufficiency, cracks in pin housing and spacer rings. | Oil leakage, loose connections. |
To model the vibration signals associated with these faults, we often use time-frequency analysis methods. For instance, the vibration acceleration signal $a(t)$ can be decomposed into components related to specific fault frequencies. The root mean square (RMS) value of vibration is a common indicator for fault severity, calculated as:
$$ a_{\text{rms}} = \sqrt{\frac{1}{T} \int_0^T a(t)^2 \, dt } $$
where $T$ is the sampling period. For gear faults, the meshing frequency $f_m$ is given by:
$$ f_m = \frac{N \cdot n}{60} $$
with $N$ being the number of teeth and $n$ the rotational speed in rpm. In an RV reducer, multiple stages lead to complex frequency components, which require advanced signal processing for isolation.
Our platform design adopts an integrated data acquisition solution that consolidates multiple performance parameter collection functions into a comprehensive system. This approach offers advantages such as low cost, small footprint, high data acquisition efficiency, and the ability to conduct various performance tests under identical working conditions. The overall scheme structure consists of a servo motor, servo motor driver, triaxial temperature-vibration intelligent sensor, RV reducer, encoder, noise spectrum analyzer, and data acquisition card. The platform surface uses a 6–8 mm thick SUS430 high magnetic permeability stainless steel plane. To enhance stability during operation, vibration damping pads are installed under the platform. The reducer support is fixed on a sliding platform that allows independent displacement in the front-back direction along rails, facilitating the assembly of different RV reducers. The pad platform’s main material is heat-treated reinforced steel to ensure structural stability under load.
The architecture diagram illustrates the connections: the servo motor is directly coupled to the input shaft of the RV reducer via a coupling, while the output shaft is rigidly connected to a fixed load or free end via another coupling. The sliding platform adjusts the axial alignment of the servo motor, RV reducer, and load end to ensure coaxiality. An industrial computer drives the servo motor according to preset speed or torque modes, and the servo driver feedbacks real-time output torque and speed parameters. During operation, the triaxial vibration-temperature sensor is directly mounted on the RV reducer housing surface, continuously collecting vibration acceleration signals in the X, Y, and Z directions along with temperature data. The noise spectrum analyzer synchronously records noise signals, extracting characteristic frequency components through frequency domain analysis. Servo motor operating parameters (e.g., current, power) are transmitted to the industrial computer via the driver bus, combined with load-end feedback to deduce the dynamic characteristics of the transmission system. All sensor data are transmitted to the computer via the data acquisition card. Through vibration spectrum analysis, temperature trend comparison, and noise frequency domain feature mining, the operating state and fault characteristics of the reducer are comprehensively assessed. Closed-loop control algorithms dynamically adjust the servo motor’s input parameters to maintain test condition stability and ensure synchronization of multi-source data.
To meet the overall design requirements of the acquisition scheme, we focus on two key aspects: constructing a multi-source sensor fusion detection system with optimized spatial layout for vibration, temperature, and noise sensors, and designing a universal reducer mounting base that adapts to different models of RV reducers while maintaining precision during exchanges.
For hardware selection, we base our choices on the RV reducer’s specifications. We select an Alpha Gear GGIRV-20E RV reducer as our test subject, with key parameters summarized in the table below.
| Parameter | Value |
|---|---|
| Rated Torque | 167 N·m |
| Start/Stop Allowable Torque | 412 N·m |
| Rated Speed | 15 rpm |
| Starting Efficiency | 80% |
| Reduction Ratio | 81 |
Based on these, we choose a Pfid DB80-02430 750W servo motor, whose parameters are listed in the following table.
| Parameter | Value |
|---|---|
| Power | 750 W |
| Rated Torque | 2.4 N·m |
| Peak Torque | 7.2 N·m |
| Peak Speed | 6000 rpm |
| Rated Speed | 3000 rpm |
Through calculation, the 750W servo motor paired with the GGIRV-20E reducer (reduction ratio 81) meets torque, speed, and power requirements with margin, ensuring a rational system design.
For data acquisition devices, we utilize a triaxial vibration-temperature intelligent sensor from VibrationTech, with specifications shown in the table.
| Parameter | Value |
|---|---|
| Model | VTall-T163E-A |
| Measurement Range | Acceleration: ±16 g, Temperature: -40°C to 125°C |
| Frequency Response | DC to 6 kHz (±3 dB) |
| Resolution | 0.488 mg/LSB |
| Sampling Frequency | 26.667 kHz |
| Operating Temperature | -40°C to 85°C |
| Protection Rating | IP67 |
This sensor attaches magnetically to the RV reducer housing and transmits data via Ethernet. For noise measurement, we employ an Aihua AWA6292 sound level meter with a measurement range of 20 dB(A) to 143 dB(A) and frequency coverage from 10 Hz to 20 kHz, capable of capturing both low-frequency vibrations and high-frequency noise from faults. For current signal acquisition, we use a Simacohua USB-3100 series data acquisition card to convert analog current signals from the servo driver into digital data.
The structural design of the RV reducer platform involves a modular mechanical approach. The sliding rail-type reducer support is made from aluminum alloy 6061 and consists of three core components: the slide rail base, indexing fixed support, and screw alignment support. These enable quick replacement and precise positioning of the RV reducer through pure mechanical linkage. The slide rail base uses a composite layered structure with linear guides as the motion center to decouple the upper and lower plates dynamically. The upper carrier plate is fixed to the slide rail with M6 screws, forming a motion unit with high load capacity and accurate displacement. The lower base plate provides stable support on the table surface, with a wide design to distribute load stress and ensure overall torsional resistance and vibration suppression during operation. The screw alignment support employs a bidirectional synchronous screw transmission mechanism with dual trapezoidal screws symmetrically arranged; handwheel-driven nuts achieve opening and closing, while V-shaped jaws with vulcanized rubber non-slip layers provide self-centering clamping with an opening range of 40–120 mm. The indexing fixed support features three sets of circular hole arrays evenly distributed around the circumference, each with holes at 22.5° intervals, using M6 bolts for radial positioning. A locking unit with circular and elliptical slots accommodates various RV reducer models for detection, enabling rapid axial locking via hole-slot engagement principles. Additionally, a motor mounting support is designed to align the servo motor shaft coaxially with the RV reducer platform shaft.
During hardware configuration and testing, we follow specific procedures for each device. For the triaxial vibration-temperature sensor, after connecting to the computer, we modify the local IP parameters of the wired Ethernet connection and use configuration software (VibrationMoni V1.14) to set the device address, IP parameters, and communication protocol. Real-time waveforms and spectrum analysis are displayed in the software. For the servo driver, we select AI1 or AI2 as the current signal output channel, define analog voltage-to-torque mapping via function codes, set filter time to 2 ms, and monitor current data in real time through internal torque command and phase current effective value parameters. For the sound level meter, we use weighting mode with Z-weighting for full-frequency noise energy analysis, set sampling rate to ≥40 kHz, apply Hanning window for FFT to reduce spectral leakage, adjust resolution bandwidth to below 10 Hz for fault characteristic frequencies, enable continuous recording with internal or expanded storage, and activate digital filtering to suppress 50 Hz power frequency interference and high-frequency electromagnetic noise. For the data acquisition card, we connect the servo driver’s current output to the AI+ and AI- pins with shielded cables, set single-channel sampling rate to 1 MS/s and range to ±10.24 V, use software triggering for precise synchronization, and configure continuous acquisition mode with pre-trigger functionality to store data before triggering.
After installing and configuring all devices, we power them on for operation. The RV reducer data acquisition platform successfully collects data from all sensors, achieving the expected results. This demonstrates the feasibility and practical value of the data acquisition system, providing technical reference for the design optimization and fault diagnosis analysis of RV reducers. The platform’s ability to synchronously acquire multi-source data and perform feature extraction lays a solid foundation for subsequent fault diagnosis model construction.
In conclusion, our work addresses the fault diagnosis needs of industrial robot RV reducers by designing and developing a multi-source data acquisition platform. Through dynamic monitoring of multi-dimensional signals such as vibration, temperature, and noise, the platform supports state assessment and fault diagnosis of RV reducers. Based on the structural characteristics and typical failure mechanisms of the RV reducer, we systematically categorize key failure modes and identify vibration, temperature, and noise signals as core parameters for fault characterization. By integrating a servo drive system, triaxial vibration-temperature sensors, and a multifunctional sound level meter, we construct a modular data acquisition platform adaptable to multiple models of RV reducers. The sliding rail support design and adjustable alignment mechanism significantly enhance device compatibility and exchange efficiency. Targeting the load characteristics of the RV reducer, we complete the selection and matching calculations for servo motors, sensors, and data acquisition cards. The multifunctional sound level meter captures non-contact noise signals, extracting fault characteristic frequencies through frequency domain analysis to complement traditional vibration signal analysis. The designed platform successfully synchronizes multi-source data acquisition and feature extraction, verifying its feasibility and reliability, thus establishing a robust basis for future fault diagnosis models. Future work will involve applying machine learning algorithms to the collected data for automated fault classification and prediction, further advancing the intelligence of RV reducer maintenance systems.
To enhance the diagnostic capabilities, we plan to incorporate advanced signal processing techniques. For example, the vibration signals can be analyzed using envelope analysis to detect early stage faults. The envelope signal $e(t)$ is obtained by demodulating the high-frequency resonance components:
$$ e(t) = |a(t) + j \mathcal{H}\{a(t)\}| $$
where $\mathcal{H}$ denotes the Hilbert transform. This helps in identifying periodic impulses caused by faults like gear tooth cracks. Additionally, temperature rise $\Delta T$ can be correlated with friction losses using the formula:
$$ \Delta T = k \cdot P_{\text{loss}} $$
with $k$ as a thermal constant and $P_{\text{loss}}$ representing power loss due to inefficiencies. For noise analysis, sound pressure level $L_p$ in decibels is given by:
$$ L_p = 20 \log_{10}\left(\frac{p}{p_0}\right) $$
where $p$ is the measured sound pressure and $p_0$ is the reference pressure (20 μPa). By integrating these metrics, our platform provides a comprehensive health assessment of the RV reducer.
The platform’s modular design allows for scalability. We can easily add more sensors, such as torque sensors or oil debris monitors, to expand data dimensions. The data acquisition card supports multiple channels, enabling simultaneous sampling from all sensors. The synchronization accuracy is critical, and we achieve this by using a common clock signal and timestamping each data packet. The time synchronization error $\Delta t$ is minimized to less than 1 ms, ensuring that correlated events across sensors are accurately aligned.
In terms of data management, we implement a database system to store historical data for trend analysis. The data volume $V$ per hour can be estimated as:
$$ V = \sum_{i=1}^{N} f_{s,i} \cdot b_i \cdot 3600 $$
where $N$ is the number of sensors, $f_{s,i}$ is the sampling frequency of sensor $i$, and $b_i$ is the bytes per sample. For our setup with vibration, temperature, and noise sensors, $V$ is approximately 2 GB per hour, which is manageable with modern storage solutions.
For fault diagnosis modeling, we extract features from the acquired data. Common features include statistical moments (mean, variance, kurtosis), frequency domain features (peak frequencies, harmonic ratios), and time-frequency features (wavelet coefficients). These features serve as inputs to classifiers like support vector machines or neural networks. The performance of a classifier can be evaluated using accuracy $Acc$:
$$ Acc = \frac{\text{Number of correct predictions}}{\text{Total predictions}} $$
With sufficient data from our platform, we aim to achieve high accuracy in fault detection and classification for the RV reducer.
Overall, this data acquisition platform represents a significant step towards predictive maintenance for industrial robots. By continuously monitoring the RV reducer’s condition, we can detect faults early, schedule maintenance proactively, and avoid costly downtime. The integration of multiple sensors and advanced analytics makes the platform a powerful tool for both research and industrial applications. As we continue to refine the system, we anticipate contributions to the reliability and efficiency of robotic systems in manufacturing environments.