In the field of multidimensional force sensing, the ability to accurately measure forces and torques along three orthogonal axes and their corresponding moments is critical for applications in robotics, aerospace, and industrial automation. The Stewart platform-based six-axis force sensor has emerged as a prominent solution due to its high stiffness, symmetry, and decoupling characteristics. However, traditional wired data acquisition systems for such sensors often face challenges related to complex wiring, maintenance, and scalability. To address these issues, I have developed a multi-node wireless data acquisition system that leverages LoRa technology for efficient, low-power, and long-range data transmission. This system enables real-time monitoring of force and torque information from multiple measurement points on a self-developed Stewart six-axis force sensor, overcoming the limitations of conventional approaches. In this article, I will detail the design, implementation, and experimental validation of this system, emphasizing the integration of hardware and software components to ensure reliable data acquisition and transmission.
The core of this work revolves around a Stewart six-axis force sensor, which consists of an upper plate, a main body structure, a PCB, a base, and a power supply battery. The sensor is designed to measure forces and torques in three-dimensional space, with loading points on the upper plate allowing for precise application of multidimensional loads. The sensor’s performance specifications include a force measurement range of -1000 N to +1000 N, a torque range of -2000 N·m to +2000 N·m, a sensitivity of approximately 1 mV/V when strain gauges are applied, a resolution of 0.005 N·m, a torsional stiffness of at least 8×10^5 N·m/rad, and a maximum stress of 930.4 MPa. These metrics highlight the sensor’s capability for high-precision applications, but achieving them requires an efficient data acquisition system that can handle multiple measurement nodes without interference.
To facilitate data acquisition from the Stewart six-axis force sensor, I designed a hardware circuit comprising four main modules: the power supply module, signal processing module, microcontroller unit (MCU), and wireless communication module. The power supply module converts a 5 V input to 3.3 V for both analog and digital components using LP2985 voltage regulators, with a 0 Ω resistor connecting the analog and digital grounds to minimize noise interference. This separation is crucial in mixed-signal circuits to prevent digital noise from affecting analog signals, ensuring accurate data acquisition. The signal processing module employs an INA326EA instrumentation amplifier to convert differential signals from strain gauge full-bridge circuits into single-ended outputs, which are then fed into the MCU for analog-to-digital conversion. This setup allows for precise measurement of strain-induced voltage changes, which correlate directly with applied forces and torques on the six-axis force sensor.
The MCU selected for this system is the C8051F350, a fully integrated mixed-signal system-on-chip that features a high-speed 8051-compatible core, a 24-bit differential ADC, and various serial interfaces such as UART and SPI. The internal ADC continuously samples the strain signals at a programmable rate, and the MCU’s 768 bytes of internal RAM are used for temporary data storage. This storage capacity is critical for managing data from multiple nodes, as it must accommodate the sampled data before wireless transmission. The wireless communication module is based on the E28-2G4T12S, which utilizes the SX1280 RF chip and LoRa modulation technology. This module offers advantages such as low power consumption, long-range communication up to 3 km, high interference immunity, and a data rate configurable from 1 kbps to 2 Mbps. The module interfaces with the MCU via UART, enabling seamless data transmission to a central host computer for further processing.
The data acquisition system supports 12 measurement nodes, each corresponding to a specific location on the Stewart six-axis force sensor where strain gauges are mounted. These nodes are strategically placed on the sensor’s truss structure at positions of maximum stress, as determined through finite element analysis simulations. For instance, in the x and z directions, stress concentrations occur at specific points, guiding the placement of strain gauges in a “top-bottom” configuration to optimize signal acquisition. Each node independently samples strain data, but to avoid channel contention in the wireless transmission, a time-division multiple access (TDMA) scheme is implemented. This approach ensures that only one node transmits data at any given time, preventing data overlap and loss.
The transmission model for the multi-node system is built around the concept of time-division multiplexing, where each node is allocated a specific time slot for data transmission. Let $T$ represent the total transmission cycle time for all nodes, $T_1$ the transmission time per node, $T_2$ the time interval between adjacent nodes, and $T_3$ the time window for control commands from the host. The relationship is given by:
$$ T = 12 \times (T_1 + T_2) + T_3 $$
The sampling frequency $f$ and the storage capacity $C$ in bytes are constrained by the MCU’s RAM and the wireless module’s buffer size. The storage requirement for sampled data is calculated as:
$$ C = T \times f \times 3 $$
where the factor 3 accounts for the 24-bit ADC output per sample. Given the MCU’s RAM limit of 768 bytes, the storage must satisfy:
$$ C = 768 \times L $$
with $L$ being the fraction of RAM used for temporary storage. Additionally, the wireless module’s maximum buffer size of 117 bytes imposes a further constraint:
$$ T \times f \times 3 \leq 108 $$
considering the data frame structure (4-byte header, 1-byte address, and 4-byte footer). The maximum data rate $V$ of the wireless module is 28,709 bytes/s, leading to the inequality:
$$ [12 \times (T_1 + T_2) + T_3] \times f \times 3 \leq V \times h \times T_1 $$
where $h$ is a redundancy factor, typically set to 0.8 in practical applications. These equations ensure that the system parameters are optimized to prevent data loss and maintain synchronization across nodes.
In practice, I set the parameters based on experimental tuning: $T_1 = 10$ ms, $T_2 = 5$ ms, and $T_3 = 50$ ms, resulting in a cycle time $T = 12 \times (10 + 5) + 50 = 230$ ms. With a sampling frequency $f = 100$ Hz, the storage requirement is $C = 0.23 \times 100 \times 3 = 69$ bytes, which is well within the MCU’s RAM limit and the wireless module’s buffer capacity. This configuration allows each node to transmit up to 23 samples per cycle, ensuring real-time data acquisition for the six-axis force sensor.
| Parameter | Value |
|---|---|
| Force Range | -1000 N to +1000 N |
| Torque Range | -2000 N·m to +2000 N·m |
| Sensitivity | 1 ± 0.2 mV/V |
| Resolution | 0.005 N·m |
| Torsional Stiffness | ≥ 8×10^5 N·m/rad |
| Maximum Stress | 930.4 MPa |
The software design for the data acquisition system is divided into lower-level firmware for the nodes and upper-level host computer software. The lower-level software, implemented on the MCU, manages data sampling, storage, and transmission based on commands from the host. The communication protocol follows a “query-response” model, where the host sends instructions such as query, start, timing, and stop commands. Upon receiving a query command, each node responds to indicate its online status. The start command initiates data sampling and storage, while the timing command synchronizes the transmission slots to correct for cumulative clock errors in the MCU’s timer. This synchronization is vital for maintaining the TDMA scheme over multiple cycles. The stop command halts all activities, placing the nodes in standby mode. The firmware ensures that data is collected continuously and transmitted in orderly time slots, with error handling mechanisms to detect and retransmit lost packets.
The host computer software is developed using WinForms and provides a graphical interface for monitoring and controlling the system. The main interface includes functional modules for sending query, start, timing, and stop commands. Users can configure the serial port and baud rate to match the wireless modules, and the interface displays real-time data from all 12 nodes, along with trend graphs for individual nodes. Data is stored in the backend for post-processing, such as calibration and decoupling analysis. This software enables users to visualize the force and torque measurements, verify system operation, and export data for further analysis. The integration of hardware and software ensures a seamless workflow from data acquisition to interpretation, enhancing the usability of the six-axis force sensor in practical applications.
To validate the system, I conducted experiments on the Stewart six-axis force sensor by applying forces and torques along each axis independently. For example, forces in the Fx direction were applied in steps from 100 N to 500 N, with 20 data samples recorded at each step. Similarly, torques in the Mx, My, and Mz directions were applied from 500 N·m to 1500 N·m in 250 N·m increments. The data from all 12 nodes was collected wirelessly and processed to decouple the multidimensional interactions. A support vector machine (SVM) model was employed for decoupling, with the 12-node measurements as inputs and the applied forces or torques as outputs. The SVM used a radial basis function (RBF) kernel, and parameters were optimized to minimize errors: C = 500, lambda = 1e-10, epsilon = 1e-6, and kernel option = 0.610.
The experimental results demonstrated the effectiveness of the data acquisition system. For instance, in the Fx direction, the measured data from the 12 nodes was used to train and test the SVM model. The table below shows a subset of the error analysis for Fx force applications, highlighting the model’s accuracy.
| Actual Force (N) | Model Output 1 (N) | Model Output 2 (N) | Model Output 3 (N) | Model Output 4 (N) | Max Absolute Error (N) |
|---|---|---|---|---|---|
| 100 | 101.34 | 101.28 | 100.69 | 104.90 | 4.90 |
| 200 | 200.34 | 199.52 | 200.34 | 200.39 | 0.48 |
| 300 | 300.33 | 300.21 | 299.93 | 300.07 | 0.39 |
| 400 | 397.80 | 398.26 | 400.17 | 400.63 | 2.20 |
| 500 | 504.91 | 500.32 | 498.28 | 500.51 | 4.91 |
The maximum full-scale error, calculated as the ratio of the maximum absolute error to the maximum applied force, was within ±1% for all directions, meeting the design specifications. The variance of the model outputs was 1.76, indicating stable and consistent performance. These results confirm that the wireless data acquisition system, combined with advanced decoupling algorithms, provides reliable and accurate measurements for the six-axis force sensor.
In conclusion, the development of this multi-node wireless data acquisition system for the Stewart six-axis force sensor represents a significant advancement in multidimensional force sensing. By integrating LoRa-based wireless communication, time-division multiplexing, and robust software control, I have created a system that overcomes the limitations of wired approaches while maintaining high precision and low power consumption. The experimental validation shows that the system achieves errors within ±1% of the full scale, making it suitable for demanding applications in robotics, aerospace, and industrial automation. Future work could focus on enhancing the system’s scalability to support more nodes, integrating machine learning for real-time decoupling, and optimizing power management for extended battery life. This research underscores the potential of wireless technologies in advancing the capabilities of six-axis force sensors, paving the way for more intelligent and flexible sensing solutions.
The mathematical models and equations used in this work are fundamental to understanding the system’s behavior. For example, the relationship between sampling frequency, storage, and transmission time can be generalized as:
$$ f_{\text{max}} = \frac{V \times h \times T_1}{3 \times [12 \times (T_1 + T_2) + T_3]} $$
where $f_{\text{max}}$ is the maximum achievable sampling frequency under the given constraints. This equation highlights the trade-offs between data rate, node count, and system resources, providing a framework for designing similar multi-node acquisition systems. Additionally, the decoupling process using SVM can be expressed as a minimization problem:
$$ \min_{w,b} \frac{1}{2} \| w \|^2 + C \sum_{i=1}^{n} \xi_i $$
subject to $y_i (w \cdot \phi(x_i) + b) \geq 1 – \xi_i$ and $\xi_i \geq 0$ for all $i$, where $x_i$ are the input vectors from the 12 nodes, $y_i$ are the applied forces or torques, $w$ is the weight vector, $b$ is the bias, and $\xi_i$ are slack variables. This formulation ensures that the six-axis force sensor’s outputs are accurately mapped to the applied loads, even in the presence of noise and cross-axis interference.
Overall, this project demonstrates the importance of a holistic approach to sensor system design, combining hardware innovation, software intelligence, and mathematical modeling. The six-axis force sensor, empowered by this wireless data acquisition system, offers a reliable solution for complex force and torque measurement tasks, contributing to the evolution of smart sensing technologies.