Full-coverage putty spraying path planning method for side wall plate of railway passenger car

Abstract

The invention discloses a full-coverage putty spraying path planning method for a side wall plate of a railway passenger car, and the method comprises the steps: obtaining the three-dimensional point cloud data of the side wall plate of the railway passenger car, extracting the information of a base plane, constructing a grid map, and placing a neuron at each grid; setting a virtual point position and an initial advancing direction of an end effector of the spraying robot; steering parameter items in a movement control equation of a biological excitation neural network algorithm are optimized, a feedback adjustment coefficient is established by utilizing topological neurons of a traversed region, and full-coverage path planning of a to-be-sprayed region is completed based on an improved biological excitation neural network; in the moving process of the end effector of the spraying robot, connectivity between regions is detected in real time through a region decomposition detection algorithm, and the optimal moving point is selected according to a risk avoidance mechanism so as to reduce the probability that the end effector of the spraying robot falls into a dead zone; according to the method, the putty spraying process requirements of existing railway vehicles are met, and the generated spraying path is coherent, orderly and low in repetition rate.

Description

Technical Field

[0001] The invention belongs to the technical field of intelligent robot control, and in particular is a method for planning a full-coverage putty spraying path for a rail passenger car side wall panel based on an improved bio-inspired neural network. Background Art

[0002] With the development of rail vehicle manufacturing industry, traditional manual spraying production methods can no longer meet the needs. Robot automatic spraying, as a new intelligent spraying technology, has gradually matured in the field of coating. However, due to the high viscosity of putty, there are no successful cases in body putty spraying. Therefore, traditional coating has always used manual scraping, using unsaturated polyester putty to scrape the entire vehicle. Even so, manual spraying and grinding will still lose 30% of the material.

[0003] Affected by the viscosity and curing reaction of putty, each time the robot spray gun is turned on, the putty will be sprayed out in the form of clumps or columns. If the spray gun is turned on and off frequently, it may even cause putty blockage. Therefore, it is necessary to reduce the number of times the spray gun is turned on and off as much as possible to reduce material waste and improve the quality of putty coating. The realization of full coverage spray path planning requires planning a collision-free continuous path from the starting position of the robot end effector to the full coverage in the specified area. The putty spraying task needs to ensure that the global path length and path overlap rate indicators of the path are optimized. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a full-coverage putty spraying path planning method for rail passenger car side wall panels in response to the above-mentioned deficiencies in the prior art. The present invention targets the actual working conditions of putty spraying and the automated spraying requirements of rail passenger cars, and utilizes an improved bio-inspired neural network algorithm to effectively reduce the path repetition rate and the total path length, so as to reduce the number of times the spray gun is opened, save putty and improve the quality of the putty coating.

[0005] In order to achieve the above technical objectives, the technical solution adopted by the present invention is:

[0006] A full-cover putty spraying path planning method for a rail passenger car side wall panel comprises:

[0007] Step 1: Obtain the 3D point cloud data of the side wall of the railway passenger car, extract the base surface information and construct a grid map. A neuron is placed at each grid and initialized.

[0008] Step 2: Introduce feedback adjustment coefficient and steering function based on sequence position, improve path point selection function, and then construct decision strategy of path optimal point, and obtain the grid unit that the robot end effector may move at the next moment according to the decision strategy of path optimal point;

[0009] Step 3, using sequence repetition value coding technology to detect the connectivity of uncovered grid cells in the grid map;

[0010] Step 4: Determine whether the uncovered grid area in the grid map is connected according to the result obtained in step 3; if not, obtain the grid unit to which the robot end effector actually moves at the next moment according to the established regional decomposition risk avoidance mechanism (i.e., steps 4.2-4.3 below); if connected, directly use the grid unit to which the robot end effector may move at the next moment calculated in step 2 as the grid unit to which the robot end effector actually moves at the next moment;

[0011] Step 5, performing spray path planning based on the full coverage spray path planning algorithm of the improved bio-inspired neural network in steps 1-4;

[0012] Step 6: When the robot end effector falls into the dead zone, the full coverage spraying path planning algorithm based on the improved bio-inspired neural network is suspended, and the point-to-point path planning algorithm of the bio-inspired neural network is used to escape until the robot end effector reaches the unsprayed area;

[0013] Step 7: Continue to execute the full coverage spray path planning algorithm based on the improved bio-inspired neural network described in step 5 until there are no uncovered grid cells in the grid map, and generate a planned path.

[0014] As a further improved technical solution of the present invention, the step 1 specifically includes:

[0015] 1.1. Use a line laser 3D scanner to collect the car body to be sprayed, obtain the point cloud data of the car body surface, and then flatten the point cloud data according to the spatial geometric information of the car body to obtain new point cloud data P;

[0016] 1.2. Determine the spray gun mist diameter σ and the coating overlap width d0, set the coating overlap width to d0 = σ / 3, and the spray point interval width of the two parallel spray paths d = σ-2d0 = σ / 3;

[0017] 1.3. Construct a global painting work map for the vehicle body. The map consists of discretized grid units. The size of each grid in the grid map is consistent. The grid map size is:

[0018]

[0019] Among them, X max (P) represents the maximum value of point cloud data P along the X axis, min (P) represents the minimum value of point cloud data P along the X axis, Y max (P) represents the maximum value of point cloud data P along the Y axis, min (P) represents the minimum value of point cloud data P along the Y axis, Indicates the number of grids in the X-axis direction. Indicates the number of grids in the Y-axis direction;

[0020] 1.4. Each grid cell in the grid map represents the surface of the vehicle body; the windows, doors and defective areas of the vehicle body are all regarded as non-spraying areas. The characterization property of the grid cell (m, n) in the non-spraying area is set to -1, that is, it is an obstacle; the characterization property of the grid cell (m, n) in the area to be sprayed is set to 0, that is, it is an uncovered grid, and the characterization property of the grid cell (m, n) that has been path-planned is set to 1, that is, it is a covered grid;

[0021] 1.5. The robot's end effector is a spray gun. It is stipulated that the path movement of the spray gun on each grid unit takes 1 unit of time to determine the entry position of the spray gun in the grid map. and the initial moving direction

[0022]

[0023] in, Indicates the grid unit where the spray gun is located at time t, t≥0, Indicates the grid unit where the spray gun is located at the initial moment, is the grid unit where the spray gun is located at the next moment. At this time, the grid unit (m,n) (0) Characteristic properties and grid cells (m,n) (1) The characterization properties of are all set to 1, which means the grid is covered;

[0024] 1.6. The grid unit is regarded as an independent neuron N, and each neuron has a topological connection relationship with the adjacent neurons;

[0025] 1.7. The change in activity value of each neuron is expressed by the shunt formula:

[0026]

[0027] Among them, x i represents the activity value of the i-th neuron, x j represents the activity value of the jth neuron in the neighborhood of the i-th neuron, A, B, and D are non-negative constants, A represents the decay rate, B represents the upper limit of the neuron activity value, and D represents the lower limit of the neuron activity value; where [I i ] - Represented as inhibitory incentive, Indicates excitatory stimulation; I i represents the external input of the i-th neuron, which is determined by the environmental information and is defined as:

[0028]

[0029] Among them, F(m i ,n i ) represents the grid unit corresponding to the i-th neuron (m i ,n i )’s characterization properties;

[0030] Function [a] + and [a] - is a linear threshold function, defined as [a] + =max{a,0} and [a] - =max{-a,0};w ij =f(d ij ), d ij represents the Euclidean distance between the i-th neuron and the j-th neuron, E is a constant value, E>>B, each neuron only calculates neurons in the interval (0, r0), r0 is defined as the receptive field; f(d ij )for:

[0031]

[0032] Where, u is the attenuation coefficient;

[0033] 1.8. Set the activity values ​​of all neurons in the grid map to 0, and use formula (3) to update the activity values ​​of neurons in all grid cells in the grid map.

[0034] As a further improved technical solution of the present invention, the step 2 specifically includes:

[0035] 2.1. Limit the spray gun to horizontal and vertical movement only, reduce the size of the neuron receptive field r0, and change the number of neighboring neurons to 4;

[0036] 2.2. Introduce the feedback adjustment coefficient, the calculation formula is as follows:

[0037]

[0038] Among them, w sj =|N s -N j | represents the sth neuron N s With the jth neuron N j The Euclidean distance between neurons F c is a positive constant used to control the feedback strength. The grid unit (m j ,n j) is the eight-neighborhood neuron set centered on x j represents the activity value of the jth neuron; x s represents the activity value of the sth neuron in the neighborhood of the jth neuron; F(m s ,n s ) represents the grid unit corresponding to the sth neuron (m s ,n s )’s characterization properties;

[0039] 2.3. When the surrounding environment information of two neurons is the same, the same activity value will be calculated, resulting in the inability to determine the optimal position; according to the starting moving direction of the spray gun Design a steering function based on the order bit:

[0040]

[0041] Among them, O c is a non-negative constant; S j is the sequential number of the jth neuron in the current neuron neighborhood set, which is defined as:

[0042]

[0043] in, Depend on Determine, and the two vectors are perpendicular, Represents the vector between the i-th neuron and the j-th neuron; the coefficient c j and Y j Substituting it into the path point selection function, we finally get the decision formula for the optimal path point:

[0044]

[0045] 2.4. Based on formula (12), the grid unit (m, n) to which the spray gun may move at the next moment is calculated: (2) .

[0046] As a further improved technical solution of the present invention, the step 3 specifically includes:

[0047] 3.1. Grid unit (m,n) (2) The characterization property of is set to -1, which means it is an obstacle;

[0048] 3.2. Search the representation properties of each row of grids in the grid map in turn. When the representation property is not covered by the grid cell, When , continue searching until a grid cell with different representation properties appears The continuous grid cells with the same representation properties retrieved by encoding are connected sequences r ik (s,e), referred to as r i k , s is the grid unit The position index on the i-th row of the grid map, e is the grid unit The position index on the i-th row of the grid map, k is the index value of the connected sequence r on the i-th row of the grid map, and the connected sequence set of the i-th row grid cells It is expressed as:

[0049]

[0050] Complete the encoding of the connected sequence of uncovered grids on the grid map, thereby obtaining the connected sequence set of all rows:

[0051]

[0052] 3.3. Connected sequence set from the first row of the grid map Start checking, when a connected sequence set that is not an empty set appears , L=i+1 is used as the starting row for the subsequent regional connectivity check; The first connected sequence r in i 1 Assign label γ, γ = 0; then, The remaining connected sequences in are assigned with the label γ, and each time a connected sequence is assigned, γ=γ+1;

[0053] 3.4. Determine whether L exceeds the number of rows in the grid map If exceeded, skip steps 3.5 to 3.8 and proceed to step 3.9; otherwise, proceed to step 3.5;

[0054] 3.5. Extract the collections one by one The connected sequence in in Representing a collection The cardinality of

[0055] 3.6. Create an empty set T i+1 ; Extract collection Every connected sequence r in i j and Connectivity check, if Description i j The coded grid cells are The encoded grid cells are not connected, then Assign label γ, and γ=γ+1; otherwise, r ij The label is added to the set T i+1 Then, the connected sequence is constructed by formula (15) Connected parent node structure tree

[0056]

[0057] Among them, Ξ(r i j ) represents the connected sequence r i j The connected parent node structure tree of i j The position index set of all connected sequences; {i+1,k} represents a connected sequence Position index; through the structure tree Can be traced back to rows 1 to i and All connected sequences that are connected;

[0058] 3.7. When the set T i+1 When the cardinality of is greater than 1, it means that there are more than two connected sequence-coded grid cells in the i-th row. There is connectivity between the encoded grid cells. In this case, the set T i+1 The minimum label value in is given Right now Then query Connected parent node structure tree Update the labels of all connected sequences included:

[0059]

[0060] Among them, S represents the connected parent node structure tree The position index set of the connected sequence in, for example, S = {H1, C1, H2, C2, ..., H n ,C n}, where C n represents the index of the connected sequence, H n Represents the index of the row where the connected sequence is located, so the connected sequence can be located Represents a connected sequence The label of; S(2t) means taking the number at the 2tth position in the set S, and S(2t-1) means taking the number at the 2t-1th position in the set S;

[0061] When the set T i+1 When the cardinality of is 1, it means that there is only one grid cell connected to the sequence code in the i-th row. The coded grid cells are connected, so we only need to convert the set T i+1 The elements in are given to

[0062] 3.8, i=i+1, then L=i+1 is used as the starting row for the subsequent regional connectivity check, and step 3.4 is executed;

[0063] 3.9. Get the label set of all connected sequences in the grid map If the number of elements in the set is not 1 after removing duplicate elements, it means that the uncovered grid cells in the grid map are not connected, and the connectivity of the grid map needs to be rechecked, that is, the numbered backpropagation is performed based on the connected parent node structure tree of the connected sequence. Specifically, a conditional judgment is made for each connected sequence r in the set R. When the number of r is not 0 and there is a connected sequence with a number of 0 in the connected parent node structure tree of r, 0 is assigned to all connected sequences in r and the connected parent node structure tree.

[0064] 3.10. Calculate the set without labels Get the number of elements in the set after repeating the element

[0065] 3.11. Grid cell (m,n) (2) The characterization property is set to 0, which means there is no coverage on the grid.

[0066] As a further improved technical solution of the present invention, the step 4 specifically includes:

[0067] 4.1. According to the calculation results of step 3.10 To determine whether the uncovered grid area in the grid map is connected, if If it is not 1, it means it is not connected, and execute steps 4.2 to 4.3;

[0068] if If it is 1, it means it is connected, then skip steps 4.2 to 4.3, and directly use the grid unit (m, n) that the spray gun may move to at the next moment calculated in step 2.4 (2) As the grid unit where the spray gun actually moves at the next moment, it is added to the path collection Grid cell (m,n) (2) The characterization property of is set to 1, which means the grid is covered;

[0069] 4.2. The grid unit (m, n) to which the spray gun may move at the next moment calculated in step 2.4 (2) , that is, grid cell (m,n) (2) The characterization property of is set to -1, which means it is an obstacle;

[0070] 4.3. Get the spray gun position at the previous moment (m,n) (1) The 4-neighborhood neuron is centered And the 4 neighboring neurons The grid units with the property of 0 in the representation form a neuron set

[0071] If the collection Not an empty set, record position (m,n) (1) The neuron activity value is x * , use formula (3) to update the activity values ​​of all neurons in the grid map, and use formula (12) to calculate the grid unit where the gun may move at the next moment The recorded neuron activity value x * Assign value to position (m,n) (1) The neuron at As the grid unit where the spray gun actually moves at the next moment, it is added to the path collection Grid Cell The characteristic property of is set to 1, which means the grid is covered, and the grid unit (m,n) (2) The characterization property of is set to 0, which means that the grid is not covered;

[0072] If the collection If it is an empty set, the grid unit (m,n) that the gun may move to at the next moment calculated in step 2.4 is (2) As the grid unit where the spray gun actually moves at the next moment, it is added to the path collection And the grid cell (m,n) (2) The characterization property of is set to 1, which means the grid is covered.

[0073] As a further improved technical solution of the present invention, the step 5 specifically includes:

[0074] 5.1. Update the activity values ​​of all neurons in the grid map using formula (3);

[0075] 5.2. Calculate the grid unit that the spray gun may move to at the next moment by formula (12);

[0076] 5.3. Calculate the connectivity of the uncovered grid areas in the grid map using steps 3.1 to 3.11;

[0077] 5.4. Use steps 4.1 to 4.3 to calculate the grid unit that the spray gun actually moves at the next moment.

[0078] The beneficial effects of the present invention are:

[0079] (1) In order to solve the problem of poor path planning in the full coverage path planning task of the bio-inspired neural network, a feedback adjustment coefficient is established based on the activity value of the attenuated neurons in the perception field of the central neuron. This can effectively solve the problem of incomplete coverage of the path in narrow areas and plan a concise and coherent path.

[0080] (2) Modifying the steering parameter in the moving path judgment formula in the original algorithm can solve the problem of the same activity value of the central neuron neighborhood and select the optimal path point.

[0081] (3) We use sequence repetition value encoding technology for grids in uncovered areas and design an algorithm for rapid detection of regional connectivity, which makes up for the weak global planning of the bio-inspired neural network algorithm. The regional decomposition avoidance mechanism established based on this algorithm greatly reduces the probability of dead zone occurrence. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 This is the plane unfolding effect diagram of the side wall panel of the railway passenger car.

[0083] Figure 2 For the construction of raster maps.

[0084] Figure 3 Schematic diagram of coating overlap cumulative deposition model and grid division.

[0085] Figure 4 N i The structure of the neighborhood neurons.

[0086] Figure 5 Encoding process for the first 4 rows of the raster map.

[0087] Figure 6 It is the label transfer process of connected sequences.

[0088] Figure 7 To improve the path planning results of bio-inspired neural networks on rail car side wall panels.

[0089] Figure 7 (a) shows the path planning result of high-resolution grid map.

[0090] Figure 7 (b) is the path planning result of low-resolution raster map.

[0091] Figure 8 To improve the path planning results of bio-inspired neural networks on complex bus side wall panels. DETAILED DESCRIPTION

[0092] The specific embodiments of the present invention are further described below according to the accompanying drawings:

[0093] A full-cover putty spraying path planning method for a rail passenger car side wall panel comprises:

[0094] Step 1: Obtain the 3D point cloud data of the side wall of the railway passenger car, extract the base surface information and construct a grid map. A neuron is placed at each grid and initialized.

[0095] Step 2: Introduce feedback adjustment coefficient and steering function based on sequence position, improve path point selection function, and then construct decision strategy of path optimal point, and obtain the grid unit that the robot end effector may move at the next moment according to the decision strategy of path optimal point;

[0096] Step 3, using sequence repetition value coding technology to detect the connectivity of uncovered grid cells in the grid map;

[0097] Step 4, establish a regional decomposition risk avoidance mechanism; determine whether the uncovered grid area in the grid map is connected according to the result obtained in step 3; if not, obtain the grid unit that the robot end effector actually moves at the next moment according to the established regional decomposition risk avoidance mechanism; if connected, directly use the grid unit that the robot end effector may move at the next moment calculated in step 2 as the grid unit that the robot end effector actually moves at the next moment;

[0098] Step 5, performing spray path planning based on the full coverage spray path planning algorithm of the improved bio-inspired neural network in steps 1-4;

[0099] Step 6, when the robot end effector falls into a dead zone (when there is no grid unit in the uncovered area to choose from in the next moving path of the robot end effector, it enters the dead zone), suspend the execution of the full coverage spraying path planning algorithm based on the improved bio-inspired neural network, and use the existing bio-inspired neural network point-to-point path planning algorithm to escape until the robot end effector reaches the unsprayed area;

[0100] Step 7: Continue to execute the full coverage spraying path planning algorithm based on the improved bio-inspired neural network described in step 5 until there are no uncovered grid cells in the grid map, and generate a planned path, such as Figure 7 Or as shown in 8.

[0101] Wherein, the step 1 specifically includes:

[0102] 1.1. Use a line laser 3D scanner to collect the car body to be sprayed. After obtaining the point cloud data of the car body surface, flatten the point cloud data according to the spatial geometric information of the car body to obtain new point cloud data P, such as Figure 1 As shown;

[0103] 1.2. Determine the spray gun mist diameter σ and coating overlap width d0, such as Figure 3 As shown, according to the coating superposition principle and putty deposition model, the coating overlap width is set to d0 = σ / 3, and the spray point interval width of the two parallel spray paths is d = σ-2d0 = σ / 3;

[0104] 1.3. Construct a global painting work map for the vehicle body (a grid map with a circle of grid cells filled at the edge to represent obstacles). The grid map is fully covered. Figure 2 The grid map of the spraying area is hidden in the figure to facilitate the subsequent path drawing and observation. The map is composed of discretized grid cells. The size of each grid in the grid map is consistent. The grid map size is:

[0105]

[0106] Among them, X max (P) represents the maximum value of point cloud data P along the X axis, min (P) represents the minimum value of point cloud data P along the X axis, Y max (P) represents the maximum value of point cloud data P along the Y axis, min (P) represents the minimum value of point cloud data P along the Y axis, Indicates the number of grids in the X-axis direction. Indicates the number of grids in the Y-axis direction;

[0107] 1.4. Each grid cell in the grid map represents the surface of the vehicle body; the windows, doors and defective areas of the vehicle body are all regarded as non-spraying areas. The characterization property F(m,n) of the grid cell (m,n) in the non-spraying area is set to -1, that is, it is an obstacle; the characterization property F(m,n) of the grid cell (m,n) in the area to be sprayed is set to 0, that is, it is an uncovered grid, and the characterization property F(m,n) of the grid cell (m,n) that has been path-planned is set to 1, that is, it is a covered grid;

[0108] 1.5. The robot's end effector is a spray gun. It is stipulated that the path movement of the spray gun on each grid unit takes 1 unit of time to determine the entry position of the spray gun in the grid map. and the initial moving direction

[0109]

[0110] in, Indicates the grid unit where the spray gun is located at time t, t≥0, Indicates the grid unit where the spray gun is located at the initial moment, is the grid unit where the spray gun is located at the next moment. At this time, the grid unit (m,n)(0) Characteristic properties and grid cells (m,n) (1) The characterization properties of are all set to 1, that is, the grid is covered; where m in the grid cell (m, n) represents the horizontal coordinate position of the grid cell, and n represents the vertical coordinate position of the grid cell.

[0111] 1.6. The grid unit is regarded as an independent neuron N, and each neuron has a topological connection relationship with the adjacent neurons;

[0112] 1.7. The change in the activity value of a neuron is affected by other neurons in the receptive field, showing excitement or inhibition. The change in the activity value of each neuron is expressed by the shunt formula:

[0113]

[0114] Among them, x i represents the activity value of the i-th neuron, x j represents the activity value of the jth neuron in the neighborhood of the i-th neuron, A, B, and D are non-negative constants, A represents the decay rate, B represents the upper limit of the neuron activity value, and D represents the lower limit of the neuron activity value; where [I i ] - Represented as inhibitory incentive, Indicates excitatory stimulation; I i represents the external input of the i-th neuron, which is determined by the environmental information and is defined as:

[0115]

[0116] Among them, F(m i ,n i ) represents the grid unit corresponding to the i-th neuron (m i ,n i )’s characterization properties;

[0117] Function [a] + and [a] - is a linear threshold function, defined as [a] + =max{a,0} and [a] - =max{-a,0};w ij =f(d ij ), d ij represents the Euclidean distance between the i-th neuron and the j-th neuron, E is a constant value, E>>B, each neuron only calculates neurons in the interval (0, r0), r0 is defined as the receptive field; f(d ij )for:

[0118]

[0119] Where, u is the attenuation coefficient;

[0120] 1.8. Set the activity values ​​of all neurons in the grid map to 0, and use formula (3) to update the activity values ​​of neurons in all grid cells in the grid map.

[0121] Wherein, the step 2 specifically includes:

[0122] 2.1. In the spraying process, if an oblique spraying path appears, the local coating thickness will increase. Therefore, in order to meet the requirements of the spraying process, the spray gun is limited to horizontal and vertical movement, and the size of the neuron receptive field r0 is reduced, and the number of judgments of the neighborhood neurons is changed to 4, such as Figure 4 As shown;

[0123] 2.2. The activity value of the neurons in the covered grid cells will gradually decrease as the time step increases, which will stimulate the neurons in the adjacent area to a certain extent; the activity of the neurons in the obstacle grid cells is a small positive value, which will be repelled and moved away during path planning. This characteristic is used to introduce the feedback adjustment coefficient, and the calculation formula is as follows:

[0124]

[0125] Among them, w sj =|N s -N j | represents the sth neuron N s With the jth neuron N j The Euclidean distance between neurons F c is a positive constant used to control the feedback strength. The grid unit (m j ,n j ) is the eight-neighborhood neuron set centered on x j represents the activity value of the jth neuron; x s represents the activity value of the sth neuron in the neighborhood of the jth neuron; F(m s ,n s ) represents the grid unit corresponding to the sth neuron (m s ,n s )’s characterization properties;

[0126] 2.3. When the surrounding environment information of two neurons is the same, the same activity value will be calculated by formula (3), resulting in the inability to determine the optimal position; according to the starting moving direction of the spray gun Design a steering function based on the order bit:

[0127]

[0128] Among them, O c is a non-negative constant; S j is the sequential number of the jth neuron in the current neuron neighborhood set, which is defined as:

[0129]

[0130] in, Depend on Determine, and the two vectors are perpendicular, Represents the vector between the i-th neuron and the j-th neuron; the coefficient c j and Y j Substitute into the path point selection function; where the path point selection function is:

[0131]

[0132] in, represents the neuron activity value at the selected moving path point, p n is the next position of the robot. c is a positive constant, y j Is a function related to mobile steering, the function is defined as:

[0133]

[0134] Among them, Δθ j Represents the current path moving direction θ c and the next possible path moving direction θ j The angle range is [0,π], and the formula is:

[0135]

[0136] Indicates the grid position of the spray gun at the current moment. Indicates the grid position of the spray gun at the next moment, (x pp ,y pp ) represents the grid position of the spray gun at the previous moment;

[0137] Finally, the decision formula for the optimal point of the path is obtained:

[0138]

[0139] 2.4. Based on formula (12), when j is 1, 2, 3, and 4 respectively, the largest neuron activity value x is selected. j +c jThe grid unit corresponding to Y is used as the moving path point where the spray gun may move at the next moment, and then the grid unit (m, n) where the spray gun may move at the next moment is obtained. (2) .

[0140] Wherein, the step 3 specifically includes:

[0141] 3.1. Grid unit (m,n) (2) The characterization property of is set to -1, which means it is an obstacle;

[0142] 3.2. Search the representation properties of each row of grids in the grid map in turn. When the representation property is not covered by the grid cell, When , continue searching until a grid cell with different representation properties appears The continuous grid cells with the same representation properties retrieved by encoding are connected sequences r i k (s,e), referred to as r i k ,like Figure 5 As shown, s is the grid unit The position index on the i-th row of the grid map, e is the grid unit The position index on the i-th row of the grid map, k is the index value of the connected sequence r on the i-th row of the grid map, and the connected sequence set of the i-th row grid cells It is expressed as:

[0143]

[0144] Complete the encoding of the connected sequence of uncovered grids on the grid map, thereby obtaining the connected sequence set of all rows:

[0145]

[0146] 3.3. Connected sequence set from the first row of the grid map Start checking, when a connected sequence set that is not an empty set appears , L=i+1 is used as the starting row for the subsequent regional connectivity check; The first connected sequence r in i 1 Assign label γ, γ = 0; then, The remaining connected sequences in are assigned with the label γ, and each time a connected sequence is assigned, γ=γ+1;

[0147] 3.4. Determine whether L exceeds the number of rows in the grid map If exceeded, skip steps 3.5 to 3.8 and proceed to step 3.9; otherwise, proceed to step 3.5;

[0148] 3.5. Extract the collections one by one The connected sequence in in Representing a collection The cardinality of

[0149] 3.6. Create an empty set T i+1 ; Extract collection Every connected sequence r in i j and Connectivity check, if (That is, if the kth connected sequence in the i+1th row There is no subset a connected to the jth sequence r in the i-th row. i j Any subset b in is equal, indicating that the two connected sequences of adjacent rows are not connected), indicating that r i j The coded grid cells are The encoded grid cells are not connected, then Assign label γ, and γ=γ+1; otherwise, r i j The label is added to the set T i+1 Then, the connected sequence is constructed by formula (15) Connected parent node structure tree

[0150]

[0151] Among them, Ξ(r i j ) represents the connected sequence r i j The connected parent node structure tree of i j The position index set of all connected sequences; {i+1,k} represents a connected sequence Position index; through the structure tree Can be traced back to rows 1 to i and All connected sequences that are connected;

[0152] 3.7. When the set T i+1 When the cardinality of is greater than 1, it means that there are more than two connected sequence-coded grid cells in the i-th row. There is connectivity between the encoded grid cells. In this case, the set T i+1 The minimum label value in is given Right now Then query Connected parent node structure tree Update the labels of all connected sequences contained, such as Figure 6 As shown:

[0153]

[0154] Among them, S represents the connected parent node structure tree The position index set of the connected sequence in, for example, S = {H1, C1, H2, C2, ..., H n ,C n}, where C n represents the index of the connected sequence, H n Represents the index of the row where the connected sequence is located, so the connected sequence can be located Represents a connected sequence The label of; S(2t) means taking the number at the 2tth position in the set S, and S(2t-1) means taking the number at the 2t-1th position in the set S;

[0155] When the set T i+1 When the cardinality of is 1, it means that there is only one grid cell connected to the sequence code in the i-th row. The coded grid cells are connected, so we only need to convert the set T i+1 The elements in are given to

[0156] 3.8, i=i+1, then L=i+1 is used as the starting row for the subsequent regional connectivity check, and step 3.4 is executed;

[0157] 3.9. Get the label set of all connected sequences in the grid map If the number of elements in the set after removing duplicate elements is not 1 (assuming At this time, removing the repeated elements will result in {0, 1, 2}, and the number of elements is 3), indicating that the uncovered grid cells in the grid map are not connected. It is necessary to recheck the connectivity of the grid map, that is, to perform label backpropagation based on the connected parent node structure tree of the connected sequence. Specifically, Make a conditional judgment for each connected sequence r in the structure tree. When the label of r is not 0 and there is a connected sequence labeled 0 in the connected parent node structure tree of r, assign 0 to all connected sequences in r and the connected parent node structure tree.

[0158] 3.10. Calculate the set without labels Get the number of elements in the set after repeating the element

[0159] 3.11. Grid cell (m,n) (2) The characterization property is set to 0, which means there is no coverage on the grid.

[0160] Wherein, the step 4 specifically includes:

[0161] 4.1. According to the calculation results of step 3.10 To determine whether the uncovered grid area in the grid map is connected, if If it is not 1, it means it is not connected, and execute steps 4.2 to 4.3;

[0162] if If it is 1, it means it is connected, then skip steps 4.2 to 4.3, and directly use the grid unit (m, n) that the spray gun may move to at the next moment calculated in step 2.4 (2) As the grid unit where the spray gun actually moves at the next moment, it is added to the path collection (If it is connected, it means that the next moment's spray gun position calculated in step 2.4 will not cause a dead zone. If there is no dead zone, it ensures that there will be no overlapping paths in the planned spray gun movement path, so it can be directly added to the path set), grid unit (m, n) (2) The characterization property of is set to 1, which means the grid is covered;

[0163] 4.2. The grid unit (m, n) to which the spray gun may move at the next moment calculated in step 2.4 (2) , that is, grid cell (m,n) (2) The characterization property of is set to -1, which means it is an obstacle;

[0164] 4.3. Get the spray gun position at the previous moment (m,n) (1) The 4-neighborhood neuron is centered And the 4 neighboring neurons The grid units with the property of 0 in the representation form a neuron set

[0165] If the collection Not an empty set, record position (m,n) (1) The neuron activity value is x * , use formula (3) to update the activity values ​​of all neurons in the grid map, and use formula (12) to calculate the grid unit where the gun may move at the next moment The recorded neuron activity value x * Assign value to position (m,n) (1) The neuron at As the grid unit where the spray gun actually moves at the next moment, it is added to the path collection Grid Cell The characteristic property of is set to 1, which means the grid is covered, and the grid unit (m,n) (2)The characterization property of is set to 0, which means that the grid is not covered;

[0166] If the collection If it is an empty set, the grid unit (m,n) that the gun may move to at the next moment calculated in step 2.4 is (2) As the grid unit where the spray gun actually moves at the next moment, it is added to the path collection And the grid cell (m,n) (2) The characterization property of is set to 1, which means the grid is covered.

[0167] Wherein, the step 5 specifically includes:

[0168] 5.1. Update the activity values ​​of all neurons in the grid map using formula (3);

[0169] 5.2. Calculate the grid unit that the spray gun may move to at the next moment by formula (12);

[0170] 5.3. Calculate the connectivity of the uncovered grid areas in the grid map using steps 3.1 to 3.11;

[0171] 5.4. Use steps 4.1 to 4.3 to calculate the grid unit that the spray gun actually moves at the next moment.

[0172] The protection scope of the present invention includes but is not limited to the above embodiments. The protection scope of the present invention shall be based on the claims. Any replacement, deformation, and improvement of the technology that can be easily thought of by technicians in this field shall fall within the protection scope of the present invention.

Claims

1. A full coverage putty spraying path planning method for a rail passenger car side wall panel, characterized in that: include: Step 1: Obtain the 3D point cloud data of the side wall of the railway passenger car, extract the base surface information and construct a grid map. A neuron is placed at each grid and initialized. Step 2: Introduce feedback adjustment coefficient and steering function based on sequence position, improve path point selection function, and then construct decision strategy of path optimal point, and obtain grid unit that the robot end effector may move at the next moment according to the decision strategy of path optimal point; Step 3, using sequence repetition value coding technology to detect the connectivity of uncovered grid cells in the grid map; Step 4: Determine whether the uncovered grid area in the grid map is connected according to the result obtained in step 3; if not, obtain the grid unit to which the robot end effector actually moves at the next moment according to the established regional decomposition risk avoidance mechanism; if connected, directly use the grid unit to which the robot end effector may move at the next moment calculated in step 2 as the grid unit to which the robot end effector actually moves at the next moment; Step 5, performing spray path planning based on the full coverage spray path planning algorithm of the improved bio-inspired neural network in steps 1-4; Step 6: When the robot end effector falls into the dead zone, the full coverage spraying path planning algorithm based on the improved bio-inspired neural network is suspended, and the point-to-point path planning algorithm of the bio-inspired neural network is used to escape until the robot end effector reaches the unsprayed area; Step 7: Continue to execute the full coverage spray path planning algorithm based on the improved bio-inspired neural network described in step 5 until there are no uncovered grid cells in the grid map, and generate a planned path.

2. The full coverage putty spraying path planning method for rail passenger car side wall panels according to claim 1 is characterized in that: The step 1 specifically includes: 1.

1. Use a line laser 3D scanner to collect the car body to be sprayed, obtain the point cloud data of the car body surface, and then flatten the point cloud data according to the spatial geometric information of the car body to obtain new point cloud data P; 1.

2. Determine the spray gun mist diameter σ and the coating overlap width d0, set the coating overlap width to d0 = σ / 3, and the spray point interval width of the two parallel spray paths d = σ-2d0 = σ / 3; 1.

3. Build a global painting work map for the vehicle body. The map consists of discretized grid units. The size of each grid in the grid map is consistent. The grid map size is: Among them, X max (P) represents the maximum value of point cloud data P along the X axis. min (P) represents the minimum value of point cloud data P along the X axis, Y max (P) represents the maximum value of point cloud data P along the Y axis, min (P) represents the minimum value of point cloud data P along the Y axis, Indicates the number of grids in the X-axis direction. Indicates the number of grids in the Y-axis direction; 1.

4. Each grid cell in the grid map represents the surface of the vehicle body; the windows, doors and defective areas of the vehicle body are all regarded as non-spraying areas. The characterization property of the grid cell (m, n) in the non-spraying area is set to -1, that is, it is an obstacle; the characterization property of the grid cell (m, n) in the area to be sprayed is set to 0, that is, it is an uncovered grid, and the characterization property of the grid cell (m, n) that has been path-planned is set to 1, that is, it is a covered grid; 1.

5. The robot's end effector is a spray gun. It is stipulated that the path movement of the spray gun on each grid unit takes 1 unit of time to determine the entry position of the spray gun in the grid map. and the initial moving direction in, Indicates the grid unit where the spray gun is located at time t, t≥0, Indicates the grid unit where the spray gun is located at the initial moment, is the grid unit where the spray gun is located at the next moment. At this time, the grid unit (m,n) (0) Characteristic properties and grid cells (m,n) (1) The characterization properties of are all set to 1, which means the grid is covered; 1.

6. The grid unit is regarded as an independent neuron N, and each neuron has a topological connection relationship with the adjacent neurons; 1.

7. The change in activity value of each neuron is expressed by the shunt formula: Among them, x i represents the activity value of the i-th neuron, x j represents the activity value of the jth neuron in the neighborhood of the i-th neuron, A, B, and D are non-negative constants, A represents the decay rate, B represents the upper limit of the neuron activity value, and D represents the lower limit of the neuron activity value; where [I i ] - Represented as inhibitory incentive, Indicates excitatory stimulation; I i represents the external input of the i-th neuron, which is determined by the environmental information and is defined as: Among them, F(m i ,n i ) represents the grid unit (m i ,n i )’s characterization properties; Function [a] + and [a] - is a linear threshold function, defined as [a] + =max{a,0} and [a] - =max{-a,0};w ij =f(d ij ), d ij represents the Euclidean distance between the i-th neuron and the j-th neuron, E is a constant value, E>>B, each neuron only calculates neurons in the interval (0, r0), r0 is defined as the receptive field; f(d ij )for: Where, u is the attenuation coefficient; 1.

8. Set the activity values ​​of all neurons in the grid map to 0, and use formula (3) to update the activity values ​​of neurons in all grid cells in the grid map.

3. The full coverage putty spraying path planning method for rail passenger car side wall panels according to claim 2 is characterized in that: The step 2 specifically includes: 2.

1. Limit the spray gun to horizontal and vertical movement only, reduce the size of the neuron receptive field r0, and change the number of neighboring neurons to 4; 2.

2. Introduce the feedback adjustment coefficient, the calculation formula is as follows: Among them, w sj =|N s -N j | represents the sth neuron N s With the jth neuron N j The Euclidean distance between neurons F c is a positive constant used to control the feedback strength. The grid unit (m j ,n j ) is the eight-neighborhood neuron set centered on x j represents the activity value of the jth neuron; x s represents the activity value of the sth neuron in the neighborhood of the jth neuron; F(m s ,n s ) represents the grid unit corresponding to the sth neuron (m s ,n s )’s characterization properties; 2.

3. When the surrounding environment information of two neurons is the same, the same activity value will be calculated, resulting in the inability to determine the optimal position; according to the starting moving direction of the spray gun Design a steering function based on the order bit: Among them, O c is a non-negative constant; S j is the sequential number of the jth neuron in the current neuron neighborhood set, which is defined as: in, Depend on Determine, and the two vectors are perpendicular, Represents the vector between the i-th neuron and the j-th neuron; the coefficient c j and Y j Substituting it into the path point selection function, we finally get the decision formula for the optimal path point: 2.

4. Based on formula (12), the grid unit (m, n) to which the spray gun may move at the next moment is calculated: (2) .

4. The full coverage putty spraying path planning method for rail passenger car side wall panels according to claim 3 is characterized in that: The step 3 specifically includes: 3.

1. Grid unit (m,n) (2) The characterization property of is set to -1, which means it is an obstacle; 3.

2. Search the representation properties of each row of grids in the grid map in turn. When the representation property is not covered by the grid cell, When , continue searching until a grid cell with different representation properties appears The continuous grid cells with the same representation properties retrieved by encoding are connected sequences Abbreviation s is the grid unit The position index on the i-th row of the grid map, e is the grid unit The position index on the i-th row of the grid map, k is the index value of the connected sequence r on the i-th row of the grid map, and the connected sequence set of the i-th row grid cells It is expressed as: Complete the encoding of the connected sequence of uncovered grids on the grid map, thereby obtaining the connected sequence set of all rows: 3.

3. Connected sequence set from the first row of the grid map Start checking, when a connected sequence set that is not an empty set appears , L=i+1 is used as the starting row for the subsequent regional connectivity check; The first connected sequence in Assign label γ, γ = 0; then, The remaining connected sequences in are assigned with the label γ, and each time a connected sequence is assigned, γ=γ+1; 3.

4. Determine whether L exceeds the number of rows in the grid map If exceeded, skip steps 3.5 to 3.8 and proceed to step 3.9; otherwise, proceed to step 3.5; 3.

5. Extract the collections one by one The connected sequence in in Representing a collection The cardinality of 3.

6. Create an empty set T i+1 ; Extract collection Every connected sequence in and Connectivity check, if illustrate The coded grid cells are The encoded grid cells are not connected, then Assign label γ, and γ=γ+1; otherwise, The label is added to the set T i+1 Then, the connected sequence is constructed by formula (15) Connected parent node structure tree in, Represents a connected sequence The connected parent node structure tree of The position index set of all connected sequences; {i+1, k} represents a connected sequence Position index; through the structure tree Can be traced back to rows 1 to i and All connected sequences that are connected; 3.

7. When the set T i+1 When the cardinality of is greater than 1, it means that there are more than two connected sequence-coded grid cells in the i-th row. There is connectivity between the encoded grid cells. In this case, the set T i+1 The minimum label value in Right now Then query Connected parent node structure tree Update the labels of all connected sequences included: Among them, S represents the connected parent node structure tree The position index set of the connected sequence in, for example, S = {H1, C1, H2, C2, …, H n ,C n }, where C n represents the index of the connected sequence, H n Represents the index of the row where the connected sequence is located, so the connected sequence can be located Represents a connected sequence The label of; S(2t) means taking the number at the 2tth position in the set S, and S(2t-1) means taking the number at the 2t-1th position in the set S; When the set T i+1 When the cardinality of is 1, it means that there is only one grid cell connected to the sequence code in the i-th row. The coded grid cells are connected, so we only need to convert the set T i+1 The elements in are given to 3.8, i=i+1, then L=i+1 is used as the starting row for the subsequent regional connectivity check, and step 3.4 is executed; 3.

9. Get the label set of all connected sequences in the grid map If the number of elements in the set is not 1 after removing duplicate elements, it means that the grid cells in the grid map are not connected. It is necessary to recheck the connectivity of the grid map, that is, to perform label backpropagation based on the connected parent node structure tree of the connected sequence. Specifically, Make a conditional judgment for each connected sequence r in the structure tree. When the label of r is not 0 and there is a connected sequence labeled 0 in the connected parent node structure tree of r, assign 0 to all connected sequences in r and the connected parent node structure tree. 3.

10. Calculate the set without labels Get the number of elements in the set after repeating the element 3.

11. Grid cell (m,n) (2) The characterization property is set to 0, which means there is no coverage on the grid.

5. The full coverage putty spraying path planning method for rail passenger car side wall panels according to claim 4 is characterized in that: Step 4 is as follows: include: 4.

1. According to the calculation results of step 3.10 To determine whether the uncovered grid area in the grid map is connected, if If it is not 1, it means it is not connected, and execute steps 4.2 to 4.3; if If it is 1, it means it is connected, then skip steps 4.2 to 4.3, and directly use the grid unit (m, n) that the spray gun may move to at the next moment calculated in step 2.4 (2) As the grid unit where the spray gun actually moves at the next moment, it is added to the path collection Grid cell (m,n) (2) The characterization property of is set to 1, which means the grid is covered; 4.

2. The grid unit (m, n) to which the spray gun may move at the next moment calculated in step 2.4 (2) , that is, grid cell (m,n) (2) The characterization property of is set to -1, which means it is an obstacle; 4.

3. Get the spray gun position at the previous moment (m,n) (1) The 4-neighborhood neuron is centered And the 4 neighboring neurons The grid units with the property of 0 in the representation form a neuron set If the collection Not an empty set, record position (m,n) (1) The neuron activity value is x * , use formula (3) to update the activity values ​​of all neurons in the grid map, and use formula (12) to calculate the grid unit where the gun may move at the next moment The recorded neuron activity value x * Assign value to position (m,n) (1) The neuron at As the grid unit where the spray gun actually moves at the next moment, it is added to the path collection Grid Cell The characteristic property of is set to 1, which means the grid is covered, and the grid unit (m,n) (2) The characterization property of is set to 0, which means that the grid is not covered; If the collection If it is an empty set, the grid unit (m,n) that the gun may move to at the next moment calculated in step 2.4 is (2) As the grid unit where the spray gun actually moves at the next moment, it is added to the path collection And the grid cell (m,n) (2) The characterization property of is set to 1, which means the grid is covered.

6. The full coverage putty spraying path planning method for railway passenger car side wall panel according to claim 5 is characterized in that: The step 5 specifically includes: 5.

1. Update the activity values ​​of all neurons in the grid map using formula (3); 5.

2. Calculate the grid unit that the spray gun may move to at the next moment by formula (12); 5.

3. Calculate the connectivity of the uncovered grid areas in the grid map using steps 3.1 to 3.11; 5.

4. Use steps 4.1 to 4.3 to calculate the grid unit that the spray gun actually moves at the next moment.

Images