Some thoughts on vibration suppression

The current situation of the 3D vision robot project In order to meet the requirements of a wide field of view and high accuracy, we currently use multi-line lasers for measurement. After generating a point cloud image, the data is compressed through triangle reconstruction, then image stitching is performed, and then imported into BIM. Control the robot's positioning, mapping, motion control and other operations.

The difficulty of the project on the multi-line laser has been demonstrated several times. I think that the multi-line laser is very feasible because of its low cost and meeting the demand. I think vibration suppression should be a problem that needs to be solved at present: it will affect our sensor positioning speed and algorithm efficiency. Take the wall panel installation robot as an example. This is a small robot handling and installing large and heavy materials. A wall panel At least 300 kilograms and a high load-to-weight ratio. Therefore, when the robot lifts the wallboard, there will be vibration problems due to inertia. We have to wait for these vibrations to end before we can use the algorithm for visual positioning, and then slowly press into the hole. This will slow down the installation rhythm and affect building efficiency. Frequent vibration will reduce the fatigue life of the machine, reduce performance and accuracy, and increase costs. Therefore, vibration suppression must be performed.

The trouble with traditional vibration suppression technology There are two types of traditional vibration suppression technology. One is to slow down the work speed, so the vibration will be small, but it will affect the efficiency. The second is to improve the mechanical rigidity, that is, to increase the grip strength of our robot, so that the vibration will be reduced, but the cost will increase, and the inertia will also increase, which is not the optimal solution. Omron's vibration suppression, the industry's best solution is servo plus algorithm, so that the cost is low, the effect is good, I will introduce in detail below.

Causes of vibration

Constant self-correction due to deviation from the course during movement Load vibration caused by the sudden stop of the robot Unbalance of the robot caused by the change of the center of gravity of the load

solution

Deviated course correction uses encoder technology. I talked to Omron engineers about this problem, and I understood it only after listening to it. This is actually similar to the frame stabilization algorithm commonly used by TV stations. For example, when we watch TV, there will be many tracks: picture, left channel, right channel, subtitles, etc. These tracks must be synchronized with key frames to be watched normally. If a frame of picture is delayed by too much data. Seconds, the other tracks remain normal. At this time, the audio, picture and subtitles will not be synchronized, which will affect the viewing experience. Therefore, when transmitting TV signals, at least one or two synchronization signals must be added, and the progress of each track should be checked and fine-tuned every tens of milliseconds to ensure that there is no confusion. The encoder also has the same principle. It also has a synchronization signal to fine-tune the position of the servo motor in real time, but he made this algorithm into hardware. The hardware is not only fast, but the hardware cable has a shielding layer, which can eliminate mutual interference and reduce the difficulty of software writing. Therefore, with the encoder, the servo motor can realize self-correction of the course and become a continuously changing system. Its vibration suppression becomes the same as the sudden stop of the robot.

The load vibration caused by the sudden stop of the robot is mainly caused by inertia, which can be reduced but not completely eliminated. The most effective method is to add a reverse and gradually attenuating high-frequency vibration wave to the load before the robot stops moving, which cancels out the vibration wave generated by inertia, so as to achieve the purpose of rapid stability. This technology is similar to the "image steady-state derotation" technology in the sensor, adding a reverse derotation angle to make the image normal. I recently found the 2010.04 Journal of Sensing Technology "Real-time Image Based on FPGA + DSP" on the Internet Derotation System", I have read this paper several times carefully. Although derotation can be achieved with gyroscope and algorithm prediction, the cost is very high. FPGA and DSP are both cheap chips, which can perform graphics, signal and Algorithm acceleration. DSP reduces the time complexity of polynomial multiplication from O(n²) to O(nlogn) through Fourier transform. The idea of ​​DSP is similar to quick sorting and merge sorting in software. FPGA achieves acceleration by reducing the CPU data bus to 8 bits, which makes it faster than the CPU and more energy-efficient than the GPU. Therefore, they perform a large number of calculations to generate the reverse derotation angle, which is not only fast, but also low in cost. The reverse frequency is also the same principle. Its specific implementation is called servo motor PID control: PID is the abbreviation of proportional integral derivative control. It monitors the motor speed and torque signal, and then recognizes the frequency when the motor is rotating, and then controls the feedback circuit to control the speed. Correct, and then measure the frequency through speed and torque again, and the cycle continues. Since vibrations are all high-frequency, these measured frequencies must be used through amplifiers. PID is the core technology in the servo motor feedback system. P represents the error ratio, I represents the integral ratio of the error, and D represents the differential ratio of the error. In engineering, PI is often used in conjunction and can adapt to more than 80% of the scenes, but because I is an integral, it has a delay effect and cannot predict the trend in advance, so it is not used much in the field of vibration suppression. D is the force on the differential term. It is equivalent to a spring. Differentiation can predict the trend, and it can also cancel or suppress in an instant, so that the oscillation can be quickly attenuated and the system can be quickly stabilized. However, the problem with the differentiation is that as the frequency increases, the gain will become larger. It is high frequency, which means that using differential vibration suppression will amplify noise, which may be counterproductive.

There are generally two methods for the imbalance of the robot caused by the change in the center of gravity of the load: one method is to install a reducer, so that the inertia of the load converted to the motor will be greatly reduced. The servo system in Japan usually requires the load/motor inertia ratio to be less than 5:1, Omron also follows this standard. I took a look at the concept of the reducer. It is not complicated. It generates a rotating magnetic field after power-on, and then generates a magnetic tension to compress the spring, which causes the brake ring to hang in the air. Stop, so as to achieve the purpose of rapid vibration suppression. Another way is to use deep learning algorithms to train the model to make the robot automatically adapt to changes in the load center of gravity. I don't know this, so I can't write it now, and I still have the opportunity to learn it again.