Cylindrical ring brush wire pressure shaping method
By using reinforcement learning algorithms and multi-threaded communication, the brush filament pressure can be quickly adjusted and optimized, solving the problem of low efficiency in robotic arm brush filament pressure testing equipment and improving shaping efficiency and system stability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING INST OF CONTROL ENG
- Filing Date
- 2023-11-30
- Publication Date
- 2026-05-22
Smart Images
Figure CN117772845B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to an intelligent adjustment method for the pressure of cylindrical ring brush filaments, which is applied to a cylindrical ring brush filament pressure adjustment device to achieve rapid adjustment of brush filament pressure and improve production efficiency. Background Technology
[0002] Conductive slip rings, as key components of spatial pointing mechanisms, are primarily used for power and signal transmission between rotating and stationary structures. The brush pressure directly affects the performance of the conductive ring, including static contact resistance, electrical noise, insulation, frictional torque, and lifespan. Therefore, during production and development, the brush pressure value needs to be strictly controlled, with a control accuracy typically within 0.02N. Existing robotic arm brush pressure testing equipment can measure brush pressure, but its efficiency in pressure shaping is low. The shaping distance cannot be intelligently learned, and the number of shaping cycles is high, affecting the brush bending lifespan. There is an urgent need to improve the implementation method of the shaping strategy and increase shaping efficiency. Summary of the Invention
[0003] The technical problem solved by this invention is to overcome the shortcomings of the prior art and provide an intelligent adjustment algorithm and implementation for the pressure of cylindrical ring brush bristles. By using a conductive slip ring automatic measurement and adjustment device for brush bristle pressure and a reinforcement learning algorithm, the brush bristle pressure can be quickly adjusted, thus solving the problem of low efficiency and repeated bending of the brush bristles caused by simply using a robotic arm for adjustment.
[0004] The technical solution of this invention is: a method for pressure shaping of cylindrical ring brush filaments, comprising:
[0005] The brush pressure at different bristle inclination angles was measured. One brush pressure corresponds to multiple shaping distances, forming multiple shaping strategies.
[0006] A two-dimensional strategy matrix is formed based on the range of brush holder shaping distance and the range of bristle pressure variation;
[0007] The reward value of the previous shaping strategy is calculated based on the brush pressure measured in the next measurement.
[0008] Based on the two-dimensional strategy matrix and reward value, establish and continuously update the value matrix and probability matrix;
[0009] After obtaining the current bristle pressure, the current probability matrix is used to randomly select a shaping distance with a selection probability value greater than the threshold from the column corresponding to the bristle pressure. This shaping distance is then used as the final shaping distance for bristle pressure shaping.
[0010] The shaping distance range and the brush pressure variation range are two sets of discrete sequences, forming two dimensions of the two-dimensional strategy matrix. The rows in the two-dimensional strategy matrix are composed of the shaping distance range sequence, and the columns are composed of the brush pressure variation range sequence. Each coordinate in the matrix corresponds to a shaping strategy.
[0011] The reward value of the previous cosmetic surgery strategy was calculated as follows:
[0012] Let F t-1 Indicates the current bristle pressure, F t Indicates the pressure of the next bristle application after reshaping, F fmax Indicates the upper limit of the target bristle pressure, F fmin Indicates the lower limit of the target brush bristle pressure, F flag Indicates the median target bristle pressure:
[0013] When F t-1 ≥F fmax hour:
[0014]
[0015] When F fmax ≥F t-1 ≥F fmin hour:
[0016]
[0017] The closer you are to the optimal bristle pressure median, the higher the bonus value.
[0018] F represents t The reward value for exceeding the lower bound of the target pressure; the greater the exceedance of the lower bound of the target pressure, the greater the penalty.
[0019] When F t-1 ≤F fmin hour:
[0020]
[0021]
[0022] The closer you are to the optimal bristle pressure median, the higher the bonus value.
[0023] F represents t The reward value is the amount of pressure exceeding the upper limit of the target pressure; the greater the pressure exceeds the upper limit, the greater the penalty.
[0024] Establish and continuously update the value matrix, including:
[0025] A value matrix is established, with each element corresponding one-to-one with an element in the two-dimensional policy matrix. Each policy has a corresponding value in the value matrix. The elements in the value matrix are iteratively calculated from the previous value and the reward value of the current policy. The initial value of each element in the value matrix is -1. When the reward value of the current policy is received, the reward value is iterated to calculate a new value matrix.
[0026] V sa (t+1)=αV sa (t)+k(1-α)R
[0027] In the formula, V sa (t+1) represents the value at the current moment, V sa (t) represents the value at the previous moment, α represents the weighting coefficient, k represents the reward / penalty coefficient, and R represents the reward value for this decision.
[0028] After calculating the new value matrix, each element in the new value matrix is processed, that is, the value of each i,j coordinate is diffused to the surrounding P interval range according to the weight in the generalization processing matrix to obtain the new processed value matrix.
[0029] V(i,j)=β(i,j)·V(s,a),i=0,1,2,3,4,2P+1,j=0,1,2,3,4,2P+1,i≠s∩j≠a;
[0030] In the formula, V(i,j) represents the value of point i and j in the value matrix, i represents the matrix row and is an integer range sequence; j represents the matrix column and is a brush pressure range sequence; V(s,a) represents any element in the value matrix, that is, the coordinates of the center point in the generalization matrix; β(i,j) represents the weighted value of point i and j in the generalization matrix.
[0031] Establish a probability matrix, with each element in the probability matrix corresponding to a value matrix. The probability matrix represents the probability that each shaping distance can be selected under each brush pressure. That is, the higher the value of the element in the value matrix, the greater the probability that the shaping distance will be selected. The sum of the probabilities in each column of the probability matrix is 1.
[0032] The probability values of all elements in the probability matrix are calculated as follows:
[0033]
[0034] In the formula, P(i,j) represents the probability of points i and j in the probability matrix; V(i,j) represents the value of points i and j in the value matrix, where i represents the matrix row and is an integer range sequence; and j represents the matrix column and is a brush pressure range sequence.
[0035] The pressure when the stylus contacts the bristles is obtained by a pressure sensor installed at the rear of the stylus, i.e., the bristle pressure.
[0036] The shaping distance is the displacement of the stylus on the bristles. The initial value is 0. When adjusting the bristles in the forward direction, the distance increases by 0.1 or 0.2 each time, and when adjusting in the reverse direction, the distance decreases by 0.1 or 0.2 each time.
[0037] The advantages of this invention compared to the prior art are:
[0038] 1) The original robotic arm brush filament pressure measurement device could only generate the shaping distance through an incremental sequence. Each brush filament required its shaping distance to increase from 0 during adjustment, rendering the previous adjustments ineffective when the initial brush filament pressure was close to the required range. However, by utilizing an intelligent adjustment algorithm, the corresponding brush filament pressure can be learned to obtain a better or even optimal shaping distance, achieving an 80% improvement in single-adjustment efficiency.
[0039] 2) Since the characteristics of brush bristles may vary from batch to batch, different bristles may produce different results even under the same bristle pressure and the same shaping strategy. This requires the intelligent algorithm to have a certain degree of robustness. In this algorithm, when selecting the shaping distance, it is required to randomly select the shaping distance with a probability greater than 5%, which can improve the overall robustness of the algorithm and meet the requirements for dealing with the characteristics of different batches of brush bristles.
[0040] 3) Multi-threading is adopted. The host computer main program communicates with the robotic arm, and the intelligent algorithm communicates with the host computer. When the intelligent algorithm overflows the shaping safety distance, the data transmission is interrupted, but the host computer program is not interrupted. After manual inspection, the intelligent algorithm program can be restored, and the data transmission of the shaping strategy can be restored, thereby improving the safety and stability of the intelligent algorithm. Attached Figure Description
[0041] Figure 1 This is a flowchart of the method of the present invention;
[0042] Figure 2 This is a schematic diagram of the robotic arm's shaping process. The stylus is currently in the positive shaping direction, and the downward distance is the shaping distance. Detailed Implementation
[0043] The method of this invention includes obtaining an integer policy, generating a policy matrix, calculating the policy reward value, updating the value matrix and probability matrix, and selecting the integer distance. Data transmission ensures stable operation of each thread. A detailed flowchart is shown below. Figure 1 .
[0044] 1) Obtaining plastic surgery strategies
[0045] like Figure 2 As shown. The brush bristle pressure measured by the stylus and the current shaping distance form a shaping strategy. The brush bristle pressure is obtained by a pressure sensor installed at the rear of the stylus when the stylus contacts the brush bristles. The shaping distance is the displacement of the stylus on the brush bristles, with an initial value of 0. It increases by 0.1 or 0.2 each time the brush bristles are adjusted in the forward direction, and decreases by 0.1 or 0.2 each time the brush bristles are adjusted in the reverse direction.
[0046] 2) Generate the policy matrix
[0047] The strategy matrix is a two-dimensional matrix, with the two dimensions being the integral range and the discrete values of the brush pressure variation range throughout the adjustment process, respectively. The discrete interval for the integral range is 0.05, and the discrete interval for the brush pressure is 0.001 or 0.01. Each discrete value corresponds to a row or column in the strategy matrix.
[0048] The shaping range is determined based on the type of brush holder. Within this range, the bristle pressure can be adjusted without damaging the brush holder. The range of bristle pressure variation should cover all pressure values that may be measured during the entire adjustment process.
[0049] 3) Calculate the strategy reward value
[0050] The reward value is calculated based on the specific values of the brush pressure before and after the shaping strategy is adopted, as follows (let F). t-1 Indicates the current bristle pressure, F t Indicates the pressure of the next bristle application after reshaping, F fmax Indicates the upper limit of the target bristle pressure, F fmin Indicates the lower limit of the target brush bristle pressure, F flag (Indicates the median target bristle pressure):
[0051] a)F t-1 ≥F fmax :
[0052]
[0053] After a single reshaping strategy is implemented, if the change in bristle pressure is greater than 0.003, the reward is 4; if the change in bristle pressure is less than 0.003, the reward is -3. In other words, the reshaping strategy rewards those with large changes in bristle pressure.
[0054]
[0055] (F t The greater the pressure exceeds the target lower bound, the greater the penalty.
[0056] b)F t-1 ≤F fmin :
[0057]
[0058] (F t F t-1 For increases less than 0.003N, the reward is increased by -3).
[0059]
[0060] (Ft The greater the pressure exceeds the target limit, the greater the penalty.
[0061] a) Brush pressure is too high, b) Brush pressure is too low. Taking case a) as an example, if the brush pressure falls exactly at the midpoint of the required pressure range after a single reshaping, the highest reward value is given; the closer the reshaping pressure is to the midpoint, the higher the reward value, but it will not exceed the midpoint reward. When the brush pressure changes abruptly from too high to too low during the execution of the strategy, it indicates that the adjustment is too drastic, and this type of reshaping strategy will be heavily penalized; the greater the deviation from the target pressure lower bound, the greater the penalty.
[0062] Conditions a and b can be superimposed. Taking the SRA45-S-2A cylindrical ring as an example, F fmax 0.102, F fmin 0.118, F flag It is 0.11.
[0063] 4) Update the value matrix and probability matrix
[0064] A value matrix is established, with each element corresponding one-to-one with an element in the two-dimensional policy matrix. Each policy has a corresponding value in the value matrix. The elements in the value matrix are iteratively calculated from the previous value and the reward value of the current policy. The initial value of each element in the value matrix is -1. When the reward value of the current policy is received, the reward value is iterated to calculate a new value matrix.
[0065] V sa (t+1)=αV sa (t)+k(1-α)R
[0066] In the formula, V sa (t+1) represents the value at the current moment, V sa (t) represents the value at the previous moment, α represents the weighting coefficient, k represents the reward / penalty coefficient, and R represents the reward value for this decision.
[0067] Then, based on Table 1, the value values around a single point in the value action matrix are updated using a generalization matrix, as shown in Equation (5). This generalization method allows surrounding points to be selected with a probability even when there is no effective data, meaning that the value near the better point is considered better in other situations, thus accelerating the learning process of the better point-near strategy. After each generalization step, the selection probability matrix is updated again.
[0068]
[0069]
[0070] Table 1 Generalization processing matrix
[0071] V(i,j)=β(i,j)·V(s,a)(i=0,1,2,3,4,5,j=0,1,2,3,4,5,i≠s∩j≠a)
[0072] In the formula, V(i,j) represents the value of point i and j in the value matrix, i represents the matrix row and is an integer range sequence; j represents the matrix column and is a brush pressure range sequence; V(s,a) represents any element in the value matrix, that is, the coordinates of the center point in the generalization matrix; β(i,j) represents the weighted value of point i and j in the generalization matrix.
[0073] The probability values for selecting all positions in the current matrix are calculated using the following formula.
[0074]
[0075] In the formula, P(i,j) represents the probability of point i,j in the (state,action) probability matrix; V(i,j) represents the value of point i,j in the (state,action) value matrix.
[0076] 5) Selection of shaping distance
[0077] Based on the received brush pressure data from the previous step, select several next shaping actions with a probability greater than 0.05 (this probability value can be increased according to the maturity of the algorithm), and then randomly select from them.
[0078] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications to the technical solutions of the present invention based on the above-disclosed technical content without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention shall fall within the protection scope of the technical solutions of the present invention.
Claims
1. A method for pressure shaping of cylindrical ring brush filaments, characterized in that, include: The brush pressure at different bristle inclination angles was measured. One brush pressure corresponds to multiple shaping distances, forming multiple shaping strategies. A two-dimensional strategy matrix is formed based on the range of brush holder shaping distance and the range of bristle pressure variation; The reward value of the previous shaping strategy is calculated based on the brush pressure measured in the next measurement. Based on the two-dimensional strategy matrix and reward value, establish and continuously update the value matrix and probability matrix; After obtaining the current bristle pressure, the current probability matrix is used to randomly select a shaping distance with a selection probability value greater than the threshold from the column corresponding to the bristle pressure. This shaping distance is then used as the final shaping distance for bristle pressure shaping.
2. The method for pressure shaping of cylindrical ring brush filaments according to claim 1, characterized in that, The shaping distance range and the brush pressure variation range are two sets of discrete sequences, forming two dimensions of the two-dimensional strategy matrix. The rows in the two-dimensional strategy matrix are composed of the shaping distance range sequence, and the columns are composed of the brush pressure variation range sequence. Each coordinate in the matrix corresponds to a shaping strategy.
3. The method for pressure shaping of cylindrical ring brush filaments according to claim 2, characterized in that, Based on the brush pressure measured in the next measurement, the reward value of the previous shaping strategy is calculated. The specific calculation process is as follows: Let F t-1 Indicates the current bristle pressure, F t Indicates the pressure of the next bristle application after reshaping, F fmax Indicates the upper limit of the target bristle pressure, F fmin Indicates the lower limit of the target brush bristle pressure, F flag Indicates the median target bristle pressure: When F t-1 ≥F fmax hour: When F fmax ≥F t-1 ≥F fmin hour: F represents t The reward value between the upper and lower limits of the target bristle pressure is higher the closer it is to the median of the optimal bristle pressure. F represents t The reward value for exceeding the lower bound of the target pressure; the greater the exceedance of the lower bound of the target pressure, the greater the penalty. When F t-1 ≤F fmin hour: The closer you are to the optimal bristle pressure median, the higher the bonus value. F represents t The reward value is the amount of pressure exceeding the upper limit of the target pressure; the greater the pressure exceeds the upper limit, the greater the penalty.
4. The method for pressure shaping of cylindrical ring brush filaments according to claim 2, characterized in that, Establish and continuously update the value matrix, including: A value matrix is established, with each element corresponding one-to-one with an element in the two-dimensional policy matrix. Each policy has a corresponding value in the value matrix. The elements in the value matrix are iteratively calculated from the previous value and the reward value of the current policy. The initial value of each element in the value matrix is -1. When the reward value of the current policy is received, the reward value is iterated to calculate a new value matrix. V sa (t+1)=αV sa (t)+k(1-α)R In the formula, V sa (t+1) represents the value at the current moment, V sa (t) represents the value at the previous moment, α represents the weighting coefficient, k represents the reward / penalty coefficient, and R represents the reward value for this decision.
5. The method for pressure shaping of cylindrical ring brush filaments according to claim 4, characterized in that, After calculating the new value matrix, each element in the new value matrix is processed, that is, the value of each i,j coordinate is diffused to the surrounding P interval range according to the weight in the generalization processing matrix to obtain the new processed value matrix. V(i,j)=β(i,j)·V(s,a),i=0,1,2,3,4,2P+1,j=0,1,2,3,4,2P+1,i≠s∩j≠a; In the formula, V(i,j) represents the value of point i and j in the value matrix, i represents the matrix row and is an integer range sequence; j represents the matrix column and is a brush pressure range sequence; V(s,a) represents any element in the value matrix, that is, the coordinates of the center point in the generalization matrix; β(i,j) represents the weighted value of point i and j in the generalization matrix.
6. The method for pressure shaping of cylindrical ring brush filaments according to claim 5, characterized in that, Establish a probability matrix, with each element in the probability matrix corresponding to a value matrix. The probability matrix represents the probability that each shaping distance can be selected under each brush pressure. That is, the higher the value of the element in the value matrix, the greater the probability that the shaping distance will be selected. The sum of the probabilities in each column of the probability matrix is 1. The probability values of all elements in the probability matrix are calculated as follows: In the formula, P(i,j) represents the probability of points i and j in the probability matrix; V(i,j) represents the value of points i and j in the value matrix, where i represents the matrix row and is an integer range sequence; and j represents the matrix column and is a brush pressure range sequence.
7. A method for pressure shaping of cylindrical ring brush filaments according to any one of claims 1-6, characterized in that, The pressure when the stylus contacts the bristles is obtained by a pressure sensor installed at the rear of the stylus, i.e., the bristle pressure.
8. A method for pressure shaping of cylindrical ring brush filaments according to any one of claims 1-6, characterized in that, The shaping distance is the displacement of the stylus on the bristles. The initial value is 0. When adjusting the bristles in the forward direction, the distance increases by 0.1 or 0.2 each time, and when adjusting in the reverse direction, the distance decreases by 0.1 or 0.2 each time.