To view the citations/information sources, please click: Yingwei Network
Flexible actuators
(Yingwei Network, December 15, 2021) Because flexible actuators have better contact flexibility, lower weight and resistance than rigid actuators, flexible actuators are becoming more popular in the field of robotics and wearable touch. The industry has explored a range of different energy sources that power flexible actuators, including fluid drives, electrostatic drives, electromagnetic drives, and thermal drives.
Due to their high energy density and low resistance, flexible fluid actuators are widely used in wearable devices, while the literature mainly explores constant fluid pressure control (CFPC) to control flexible actuators. This control method uses a simple drive structure (one discrete valve per actuator) and a binary (on-off) control mechanism, allowing the actuator pressure to track the fluid source pressure under dynamic load.
However, CFPC therefore has several limitations: (1) limited dynamic range: the dynamic range of the force provided by the flexible actuator is limited and controlled only by its working pressure, because the actuator pressure can only switch between source pressure and atmospheric pressure; (2) the actuator response is slow: the inflation and deflation response time of the flexible actuator is determined by the source pressure and cannot be adjusted independently. (3) Low voltage control resolution: This type of control can only realize the binary control of the actuator pressure. (4) Unnatural haptic interaction: It is only possible to achieve active or exogenous control of actuator pressure, which is not intrinsically related to the interaction mode of the user and the actuator, resulting in unnatural haptic interaction.
To address the limitations, Northwestern University and Meta proposed a constant fluid quality control (CFMC) in a paper called Constant Fluidic Mass Control for Soft Actuators Using Artificial Neural Network Algorithm, in which a constant mass fluid trapped inside an actuator. When the user interacts with the actuator, any change in the actuator pressure due to this interaction can further help improve the haptic interaction. Unlike the literature that uses pressure sensors and first-principle models to estimate fluid mass, the team regulates fluid mass by precisely controlling the timing of the valve and trapping it inside the actuator.
Analog pressure control using CFMC requires a reliable model of the entire jet system, including actuators. Other researchers have developed theoretical models based on first-principles to predict actuator behavior, but the model can only approximate the actuator's behavior in a limited set of inputs, and cannot reliably capture all the nonlinearities in the fluid system, making it difficult to generalize on different flexible actuators. In the study, Northwestern University and Meta proposed a new fluid-driven solution, CFMC, for wearable haptics.
According to reports, this method allows for greater dynamic range, faster response times and simulated pressure control of flexible actuators, and produces more natural tactile interactions. The researchers implemented a jet system to demonstrate the CFMC method and experimented with the flexible actuator using the system. In addition, the team proposed a neural network-based supervised learning algorithm that enables a flexible actuator simulating pressure control to use CFMC and generalize to a new actuator.
Experiments compared CFMC with CFPC methods. The results show that CFMC can increase the dynamic range of the flexible actuator, shorten its response time to reach the desired pressure, and achieve simulated pressure control to provide more natural haptic feedback.
Figure 1 shows the jet implementation of CFPC and CFMC. The fundamental difference between these two embodiments is that, in CFPC, the actuator pressure can only exist in the source or atmospheric pressure (ATM) state, while in cfMC, the actuator can be disconnected from the source and atmosphere, and therefore remain different from the source and atmosphere.
The CFPC uses a three-way valve that connects the actuator to a compressed air source via a regulating valve. CfMC, on the other hand, uses two three-way valves: one supply valve controls the intake of compressed air and the other exhaust valve controls the exhaust to the ATM. CfMC then adjusts the fluid mass inside the actuator by varying the on/off time of both valves based on the fluid source pressure and the initial actuator pressure. The CFMC can be implemented simultaneously using two bidirectional valves. For the choice of three-way valve implementation, because of the availability of the valve.
Figure 2 shows the schematic of the CFMC. The initial pressure and volume of the flexible airbag are expressed as P0 and V0, respectively. When the fingers interact with the flexible balloons, the pressure and volume of the flexible balloons become P1 and V1, respectively. Assuming that the system follows the law of ideal gas under isothermal conditions, this means that a closed system:
P0 × V0 = P1 × V1
P0 × V0 = (P0 + P) × (V1 + V )
Assuming g P × V ≈ 0, Equation 2 can be simplified to:
P = V × P0/V0
In CFPC, when the finger interacts with the flexible airbag, the volume of the flexible balloon is reduced to V1. As a result of this interaction (Equation 3), the pressure begins to increase and the actuator's fluid path remains open to the constant pressure fluid source, allowing the actuator pressure to remain constant at source pressure (P0). In CFMC, when the finger interacts with the flexible balloon, the volume of the flexible balloon decreases, resulting in an increase in air pressure (Equation 3). However, since the air is trapped inside the actuator and the fluid path from the actuator to the regulator is blocked, the increase in pressure is proportional to the decrease in volume and inversely proportional to the volume of the initial actuator.
When the difference in contact area between CFPC and CFMC is ignored, the difference in force between CFPC and CFMC will depend on the difference in the corresponding actuator pressure. Therefore, the dynamic range of forces achievable using CFMC is greater than that of CFPC.
Another advantage of CFMC is that it allows the source pressure to be higher than the actuator's operating pressure, which enables the actuator to be inflated shorter. In CFPC, on the other hand, the source pressure is determined by the operating pressure of the actuator. In addition, unlike CFPC, CFPC only allows binary control of actuator pressure, while CFMC allows more accurate analog pressure control of flexible actuators.
However, because the capacitance of the flexible actuator is often nonlinear, and many other nonlinearities exist in the fluid system, it is difficult to capture in a generalized way with first-principles models, so the researchers used a neural network-based supervised learning algorithm to implement simulated pressure control using CFMC.
For example, the quadratic relationship between the differential pressure and the flow rate of a pneumatic capacitor with a compressible fluid, the contribution of the nonlinear force displacement behavior of the actuator material to the fluid capacitance, the distributed resistance and capacitance of the fluid pipeline, and the flow blockage under the high voltage difference are certain nonlinear phenomena present in the pneumatic system.
To compare CFMC and CFPC, the researchers performed two experiments using the same experimental setup. Experiment 1 compares the dynamic range of actuator pressure between CFMC and CFPC and the reaction force generated by the actuator during interaction. Experiment 2 studied the inflation/deflation response time and simulated pressure control of a flexible actuator based on cfMC learning neural network model.
The team used two custom soft fluid actuators with different capacitances in the experiment (Figure 6), which represents the various capacitances we will see in most haptic applications. The actuator was prototyped from a heat-sealed nylon backed TPU fabric and used to develop inflatable airbags. Experiments 1 and 2 used a flexible actuator with a small capacitance (Figure 6a) and a flexible actuator with a larger capacitance (Figure 6b) to test the versatility of the proposed control method.
This study demonstrates the multiple advantages of CFMC over CFPC, but the said advantages come at the expense of additional control infrastructure. To precisely control actuator air pressure control, the team needed additional equipment, such as equipped with additional valves and pressure sensors, and more complex control strategies. In addition, because CFMC is capable of achieving a greater dynamic range, a flexible actuator needs to be designed to handle a larger pressure range.
Overall, the researchers have proposed a new flexible fluid actuator control method, CFMC, which uses the arrangement of two valves to maintain a constant fluid mass within the actuator, and it has a potential advantage in wearable touch. The research analysis shows that CFMC makes the actuator output force have a greater dynamic range and faster response time to inflation and deflation compared to CFPC.
In addition, the team demonstrated an implementation of CFMC-based simulated pressure control, using an algorithm based on a learning neural network that allows precise adjustment of actuator pressure. Experiments have also shown that the method can be generalized to different TPU fabric fluid drives.
Related Papers:
Constant Fluidic Mass Control for Soft Actuators Using Artificial Neural Network Algorithm
https://paper.nweon.com/11475
The team concluded by adding that the external interaction force of a flexible actuator is a common situation in wearable haptic applications, and that different forces acting on the actuator affect its equivalent capacitance, reducing the accuracy of learning neural network models. In the future, the researchers plan to extend the proposed neural network model by introducing preloads as the fourth input variable, and explore the architecture of recurrent neural networks to include input variables with temporal dynamic behavior in the control system. In addition, the team plans to explore the simulated pressure control of CFMC-based wearable haptic devices, and plans to examine the impact of CFMC in creating a sense of stiffness and haptics, as well as the performance of closed-loop control.