CN108857600B - A photocatalytic-based cobalt-based alloy processing method and processing platform - Google Patents

Abstract

A cobalt-based alloy processing method and processing platform based on photocatalysis. The processing method includes the optimization and adjustment of the chemical reaction process and the optimization and adjustment of the polishing path of the mechanical arm. The optimization and adjustment of the chemical reaction process is based on the linear regression algorithm established for different processing objects The linear regression model seeks conditions such as optimal light intensity, chemical additive ratio, and temperature; the optimization and adjustment of the polishing path of the mechanical arm uses a strategy network to adjust the polishing trajectory; the processing platform includes: a robot module, a polishing workbench, and a polishing liquid Add devices, IPCs, and storage. The beneficial effects of the present invention are: effectively solving the problem that the intensity of ultraviolet light and the amount of chemical additives in the chemical reaction process in the photocatalytic processing process are not easy to determine, and can effectively improve the polishing quality; training mechanical arm polishing for different processing objects The path can effectively solve the trajectory error caused by the flexibility of the polishing disc, and improve the quality and efficiency of polishing.

Description

A photocatalytic-based cobalt-based alloy processing method and processing platform

technical field

The invention relates to a photocatalytic-based cobalt-based alloy processing method and a processing platform.

Background technique

As a self-fluxing alloy material, cobalt-based alloys can be used to improve the surface properties of metal materials and form coatings with high hardness, high temperature resistance and wear resistance. This method is widely used in mold manufacturing, automobile manufacturing industry, steam turbine manufacturing and other industries. The cobalt-based alloy coating of the mold not only has the complexity of the free-form surface, but also encounters new problems such as high hardness and high wear resistance, which makes precision machining more difficult, and it is difficult to remove it quickly by ordinary mechanical processing methods. If it can be removed by specific chemical methods , making such difficult-to-process materials into easy-to-process materials can achieve the purpose of improving processing efficiency. According to the material characteristics of the cobalt element in the cobalt-based alloy, an effective reaction solvent is configured to make the cobalt atoms on the surface react to form products that are easily removed, and combined with ultraviolet photocatalysis, the relationship between photon and quantum yield is established, and the cobalt element atom is adjusted The process of bond breaking and energy release enables the reaction to be controllable. Since the cobalt-based alloy is a multi-element alloy material, it is necessary to cooperate with mechanical action to achieve material removal of insoluble products.

The material removal process of the processing platform is a coordinated process of chemical reaction, optical control and mechanical removal. Therefore, the processing technology has a great influence on the processing results. In order to optimize the processing technology, the ability of artificial intelligence technology in continuous decision-making and finding the best point is used to combine artificial intelligence technology with the photocatalytic robot processing platform to improve the photocatalytic robot. Polishing processing efficiency and processing quality.

Contents of the invention

The purpose of the present invention is to provide a photocatalytic processing platform based on artificial intelligence technology, the processing platform can select the optimal chemical reaction conditions for different processing objects, and solve the problems caused by flexible polishing discs when polishing different processing objects. The problem of polishing path error. The processing platform can adjust the processing technology for different processing objects based on artificial intelligence technology, including: optimizing and adjusting the chemical reaction process based on the linear regression model and optimizing and adjusting the polishing path of the mechanical arm based on the strategy network model.

A photocatalytic-based cobalt-based alloy processing method according to the present invention is characterized in that it comprises the following steps:

(1) Optimization and adjustment of chemical reaction process;

Step 1) For the processing object, the industrial computer operating system invokes the number m of samples of this type of processing object stored in the database in the memory, and the unit is one, and the processing evaluation index Q, the unit is mm/s; wherein the number m of processing object samples corresponds to The chemical reaction materials and conditions include basic yellow 40 solvent concentration C1, the unit is g/L, _TiO2 concentration C2, the unit is g/L, ultraviolet light intensity I, the unit is lx and temperature T, the unit is ℃; processing evaluation The index Q, the unit is mm/s, includes the machining accuracy a, the unit is mm, the surface roughness r, the unit is μm, and the processing time h, the unit is s. The larger the Q value, the better the processing quality. The mathematical formula is as follows:

Step 2) Establish the linear regression model function formula Q' as follows, w ₀ , w ₁ , w ₂ , and w ₃ are randomly initialized weight parameters:

Q'＝w ₀ +w ₁ ×C1+w ₂ ×C2+w ₃ ×I+w ₄ ×T (2);

Step 3) set up feature vector x, x is by basic yellow 40 solvent concentration C1, TiO ₂ concentration C2, ultraviolet light intensity I, the vector that temperature T forms, x=[C1, C2, I, T]; Set up loss function , the loss value L ₁ is used to measure the difference between the linear regression model output Q' obtained by the formula (2) and the actual processing evaluation index Q obtained by the formula (1), the result formula is as follows, Q ⁱ and Q' ⁱ respectively For the i-th processing the actual evaluation index value and the predicted evaluation index value:

一共更新1000次也就是需要计算1000次公式(4)，每一次都需将上一次计算得到的/>

四个权重参数代回公式(4)计算这一轮更新后的/>

Step 4) Use the stochastic batch gradient descent algorithm to iteratively update the weight value until the specified number of iterations (preset 1000 times) is reached, calculate the weight parameter value when the loss value is the smallest, and use the following iterative formula to update the weight value, α is the learning The rate is generally 0.001, t represents the iteration round, t is set to 0 at the beginning of the iteration, and the random number of Gaussian distribution generated by the program is initialized in advance.

A total of 1000 updates means that formula (4) needs to be calculated 1000 times, and each time you need to calculate the result from the previous calculation />

The four weight parameters are substituted into formula (4) to calculate this round of updated />

均代表迭代计算；in

Both represent iterative calculations;

就是对应的w₀、w₁、w₂、w₃权重值代入公式(2)就可以建立出预测模型，通过预测模型就可以预测出不同碱性黄40溶剂浓度，TiO2浓度，紫外光照强度，温度下加工评价指标的值；Step 5) After the iteration is complete

That is, the corresponding weight values of w ₀ , w ₁ , w ₂ , and w ₃ can be substituted into the formula (2) to establish a prediction model. Through the prediction model, it is possible to predict the concentration of different basic yellow 40 solvents, the concentration of TiO2, and the intensity of ultraviolet light. The value of the processing evaluation index at the temperature;

Step 6) When the Q' value is the largest according to the predictive model, that is, when the processing evaluation index is the highest, the basic yellow 40 solvent concentration C1, the _TiO2 concentration C2, the ultraviolet light intensity L, and the values corresponding to the temperature T; The operating system commands the parameters and communicates them to the UV lamp and polishing fluid adding device, so as to achieve the purpose of optimizing and adjusting the chemical reaction process;

(2) Optimizing and adjusting the polishing path of the robotic arm;

Step 1. For the processing object, the operating system of the industrial computer calls the expected trajectory y _k stored in the cloud server database, which is composed of k coordinate points. The actual trajectory during the strategy network training process is y' _k , and the error function e _k is established. The formula is as follows :

e _k ＝y _k -y' _k (5)

Through the training of the strategy network model, the value of e _k is made as small as possible, that is, the error of the polishing trajectory is reduced;

Step ② Set the reward value R according to the error value. The smaller the error is, the larger the reward value is. In order to allow the policy network to better understand the future and potential rewards, the future attenuated rewards are used here, assuming that the action is the second, and R _k is the second The reward obtained K times, its formula is as follows:

R＝R ₁ +γR ₂ +γ ² R ₃ +...+γ ^k-1 R _k (γ is the attenuation coefficient and takes 0.99) (6);

Step ③ Construct the neural network model, the input is the current coordinate value of the manipulator, and the output is the probability value matrix p=[p ₁ ,p ₂ ,p ₃ ,...,p _n ](n determined by the number of orientations set), which can be used to select actions;

Step ④ establishes the loss function L ₂ , using the cross-entropy loss function, the formula is as follows:

L ₂ =-(r*log p)/n (7)

Increases the probability of actions with higher reward values and decreases the probability of actions with lower reward values;

In step ⑤, the loss function is substituted, and the weight value in the neural network is iteratively updated using the gradient descent algorithm, and finally a policy network that can provide the optimal policy is obtained. Based on the policy network, the polishing path of the manipulator can be optimized and adjusted.

According to the processing platform constructed based on the photocatalytic cobalt-based alloy processing method of the present invention, it is characterized in that: comprising

The robot module includes an industrial robot and a flexible pneumatic polishing device. The flexible polishing device is installed on the mechanical arm of the industrial robot to adjust the position of the workpiece;

The polishing workbench includes a polishing disc for holding the polishing liquid and an illuminating device for irradiating the polishing liquid and the workpiece, the illuminating device is installed in the polishing disc, and the illuminating head of the illuminating device is aligned with the polishing liquid;

The polishing liquid adding device is arranged next to the polishing workbench, and its liquid filling port is located directly above the polishing pool of the polishing workbench, and is used for adding polishing liquid to the polishing workbench;

The industrial computer is electrically or signally connected to the control terminal of the robot module, polishing workbench and polishing liquid adding device, and is used to control the work of the robot module, polishing workbench and polishing liquid adding device, to realize the optimal adjustment of the chemical reaction process and the adjustment of the mechanical arm Polishing path optimization adjustment;

And a memory, the signal transmission end of which is bidirectionally signal-connected or electrically connected to the signal transmission port of the industrial computer through a wire or a wireless network for storing processing information.

The bottom of the polishing disc is fixedly connected with the rotary table driven by the rotating motor to realize the rotation of the polishing disc around its vertical central axis; the polishing pool is sequentially provided with an elastic layer for supporting abrasive grains, an elastic film layer and a The abrasive grain layer used for processing the mold surface, wherein the bottom end of the elastic layer is connected with the inner bottom plate of the polishing pool, the upper surface of the elastic layer is laid with an elastic film layer, and the edge of the elastic film layer is sealed and connected with the inner wall of the polishing pool, so that the elastic film The layer and the inner wall of the polishing pool form a processing area for containing the polishing liquid; the abrasive grain layer is laid on the surface of the elastic film layer, and the abrasive grain layer is filled with polishing liquid for lubrication and heat dissipation; the control terminal of the rotating motor and the main controller Corresponding control terminals are electrically connected.

The elastic layer is an array of conical helical springs, the array of conical helical springs includes a number of mutually independent conical helical springs, the small-diameter ends of the conical helical springs are affixed to the inner bottom surface of the polishing pool, so The large-diameter end of the conical coil spring is connected with a hemispherical bracket, and the arc curve of the hemispherical bracket is in contact with the lower surface of the elastic layer; the upper end of the conical coil spring is against the lower surface of the elastic film layer.

The illumination device is an ultraviolet lamp, which is fixed at the center of the rotary table, wherein the ultraviolet lamp is always irradiated on the workpiece processing area.

The optimization of the chemical reaction process adopts a linear regression algorithm for adjustment. The optimization of the polishing path of the manipulator uses a strategy network to adjust the polishing trajectory.

The data stored in the database of the memory is based on historical machining data.

The model and parameters of the neural network in the policy network are adjusted according to the actual processing situation.

The beneficial effect of the present invention is that: the present invention combines the artificial intelligence technology and the photocatalytic robot processing platform, effectively solves the problem that the ultraviolet light intensity and the amount of chemical additives in the chemical reaction process in the photocatalytic processing process are not easy to determine, and can effectively Improve the polishing quality; training the polishing path of the robotic arm for different processing objects can effectively solve the trajectory error caused by the flexibility of the polishing disc and improve the quality and efficiency of polishing.

Description of drawings

Fig. 1 is a schematic flow chart of the chemical reaction process optimization adjustment of the present invention.

FIG. 2 is a schematic diagram of the optimization and adjustment flow of the polishing path of the robotic arm in the present invention.

Fig. 3 is the platform structure of the photocatalytic processing robot of the present invention.

Detailed ways

Further illustrate the present invention below in conjunction with accompanying drawing.

Referring to the attached picture:

Embodiment 1 A photocatalytic-based cobalt-based alloy processing method according to the present invention is characterized in that it comprises the following steps:

(1) Optimize the parameters required for chemical reactions;

Step 1) For the processing object, the industrial computer operating system transfers the number m (pieces) of this type of processing object samples and the processing evaluation index Q (mm/s) stored in the database in the memory; wherein the number m (pieces) of processing object samples Corresponding chemical reaction materials and conditions include basic yellow 40 solvent concentration C1 (g/L), _TiO2 concentration C2 (g/L), ultraviolet light intensity I (lx) and temperature T (°C); processing evaluation index Q ( mm/s) includes processing accuracy a (mm), surface roughness r (μm) and processing time h (s). The larger the Q value, the better the processing quality. The mathematical formula is as follows:

Step 2) Establish the linear regression model function formula Q' as follows, w ₀ , w ₁ , w ₂ , and w ₃ are randomly initialized weight parameters:

Q'＝w ₀ +w ₁ ×C1+w ₂ ×C2+w ₃ ×I+w ₄ ×T (2);

Step 3) set up feature vector x, x is by basic yellow 40 solvent concentration C1, TiO ₂ concentration C2, ultraviolet light intensity I, the vector that temperature T forms, x=[C1, C2, I, T]; Set up loss function , the loss value L ₁ is used to measure the difference between the model output Q' (obtained by formula (2)) and the actual Q (obtained by formula (1)), the result formula is as follows, Q ⁱ and Q' ⁱ are the i-th processing The actual evaluation index value and the predicted evaluation index value.

一共更新1000次也就是需要计算1000次公式(4)，每一次都需将上一次计算得到的/>

四个权重参数代回公式(4)计算这一轮更新后的/>

Step 4) Use the stochastic batch gradient descent algorithm to iteratively update the weight value until the specified number of iterations (preset 1000 times) is reached, calculate the weight parameter value when the loss value is the smallest, and use the following iterative formula to update the weight value, α is the learning The rate is generally 0.001, t represents the iteration round, t is set to 0 at the beginning of the iteration, and the random number of Gaussian distribution generated by the program is initialized in advance.

A total of 1000 updates means that formula (4) needs to be calculated 1000 times, and each time you need to calculate the result from the previous calculation />

The four weight parameters are substituted into formula (4) to calculate this round of updated />

均代表迭代计算；in

Both represent iterative calculations;

就是对应的w₀、w₁、w₂、w₃权重值代入公式(2)就可以建立出预测模型，通过预测模型就可以预测出不同碱性黄40溶剂浓度，TiO2浓度，紫外光照强度，温度下加工评价指标的值；Step 5) After the iteration is complete

That is, the corresponding weight values of w ₀ , w ₁ , w ₂ , and w ₃ can be substituted into the formula (2) to establish a prediction model. Through the prediction model, it is possible to predict the concentration of different basic yellow 40 solvents, the concentration of TiO2, and the intensity of ultraviolet light. The value of the processing evaluation index at the temperature;

Step 6) When the Q' value is the largest according to the prediction model, that is, when the processing evaluation index is the highest, the values corresponding to the concentration of Basic Yellow 40 solvent C1, the concentration of TiO2, the intensity of ultraviolet light L, and the temperature T are obtained respectively; The system commands the parameters and communicates them to the UV lamp and the polishing liquid adding device, so as to achieve the purpose of optimizing and adjusting the chemical reaction process;

(2) Optimizing and adjusting the polishing path of the robotic arm;

Step 1. For the processing object, the operating system of the industrial computer calls the expected trajectory y _k stored in the cloud server database, which is composed of k coordinate points. The actual trajectory during the strategy network training process is y' _k , and the error function e _k is established. The formula is as follows :

e _k ＝y _k -y' _k (5)

Through the training of the strategy network model, the value of e _k is made as small as possible, that is, the error of the polishing trajectory is reduced;

Step ② Set the reward value R according to the error value. The smaller the error is, the larger the reward value is. In order to allow the policy network to better understand the future and potential rewards, the future attenuated rewards are used here, assuming that the action is the second, and R _k is the second The reward obtained K times, its formula is as follows:

R＝R ₁ +γR ₂ +γ ² R ₃ +...+γ ^k-1 R _k (γ is the attenuation coefficient and takes 0.99) (6);

Step ③ Construct the neural network model, the input is the current coordinate value of the manipulator, and the output is the probability value matrix p=[p ₁ ,p ₂ ,p ₃ ,...,p _n ](n determined by the number of orientations set), which can be used to select actions;

Step ④ Establish a loss function, using the cross-entropy loss function, the formula is as follows:

L ₂ =-(r*log p)/n (7)

Increases the probability of actions with higher reward values and decreases the probability of actions with lower reward values;

In step ⑤, the loss function is substituted, and the weight value in the neural network is iteratively updated using the gradient descent algorithm, and finally a policy network that can provide the optimal policy is obtained. Based on the policy network, the polishing path of the manipulator can be optimized and adjusted.

Embodiment 2 The processing platform constructed according to the cobalt-based alloy processing method based on photocatalysis of the present invention includes

The robot module 1 includes an industrial robot and a flexible pneumatic polishing device, and the flexible polishing device is installed on the mechanical arm of the industrial robot to adjust the position of the workpiece;

The polishing workbench 2 includes a polishing disc for containing the polishing liquid and an illuminating device for irradiating the polishing liquid and the workpiece. The illuminating device is installed in the polishing disc, and the irradiation head of the illuminating device is aligned with the polishing liquid;

The polishing liquid adding device 3 is arranged beside the polishing workbench, and its liquid filling port is located directly above the polishing pool of the polishing workbench, and is used for adding polishing liquid to the polishing workbench;

The industrial computer 4 is electrically or signally connected to the control terminal of the robot module, the polishing workbench and the polishing liquid adding device, and is used to control the work of the robot module, the polishing workbench and the polishing liquid adding device, so as to realize the optimization adjustment of the chemical reaction process and the mechanical adjustment. Optimal adjustment of arm polishing path;

And the memory 5, the signal transmission end of which is bidirectionally signal-connected or electrically connected to the signal transmission port of the industrial computer through a wire or a wireless network for storing processing information.

The bottom of the polishing disc is fixedly connected with the rotary table driven by the rotating motor to realize the rotation of the polishing disc around its vertical central axis; the polishing pool is sequentially provided with an elastic layer for supporting abrasive grains, an elastic film layer and a The abrasive grain layer used for processing the mold surface, wherein the bottom end of the elastic layer is connected with the inner bottom plate of the polishing pool, the upper surface of the elastic layer is laid with an elastic film layer, and the edge of the elastic film layer is sealed and connected with the inner wall of the polishing pool, so that the elastic film The layer and the inner wall of the polishing pool form a processing area for containing the polishing liquid; the abrasive grain layer is laid on the surface of the elastic film layer, and the abrasive grain layer is filled with polishing liquid for lubrication and heat dissipation; the control terminal of the rotating motor and the main controller Corresponding control terminals are electrically connected.

The elastic layer is an array of conical helical springs, the array of conical helical springs includes a number of mutually independent conical helical springs, the small-diameter ends of the conical helical springs are affixed to the inner bottom surface of the polishing pool, so The large-diameter end of the conical coil spring is connected with a hemispherical bracket, and the arc curve of the hemispherical bracket is in contact with the lower surface of the elastic layer; the upper end of the conical coil spring is against the lower surface of the elastic film layer.

The illumination device is an ultraviolet lamp, which is fixed at the center of the rotary table, wherein the ultraviolet lamp is always irradiated on the workpiece processing area.

The optimization of the chemical reaction process adopts a linear regression algorithm for adjustment. The optimization of the polishing path of the manipulator uses a strategy network to adjust the polishing trajectory.

The data stored in the database of the memory is based on historical machining data.

The model and parameters of the neural network in the policy network are adjusted according to the actual processing situation.

The formula of described polishing fluid is: the composition of polishing fluid comprises Basic Yellow 40 solvent and _TiO Solvent and water, wherein the mass concentration of Basic Yellow 40 is 80g/L～100g/L; Described _TiO The mass concentration is 120g/L～150g/L. Each test uses different concentrations of Basic Yellow 40 and TiO ₂ to prepare polishing fluid.

Implementation Case 3

The basic yellow 40 of different concentrations and TiO ₂ solvents are stored in the polishing liquid adding device, the outflow of the solvent is controlled by the solenoid valve, and the outflow of the solvent is randomly controlled for each processing test and the concentration of the used solvent is calculated and recorded, mixed After that, it is evenly covered on the surface of the workpiece, and then polished by the mechanical arm.

The solvent concentration data will be transmitted to the database of the server for storage so as to be recalled later.

The content described in the embodiments of this specification is only an enumeration of the implementation forms of the inventive concept. The protection scope of the present invention should not be regarded as limited to the specific forms stated in the embodiments. The protection scope of the present invention also includes those skilled in the art. Equivalent technical means conceivable according to the concept of the present invention.

Claims (5)

1. A cobalt-based alloy processing method based on photocatalysis, is characterized in that, comprises the following steps: (1) Optimization and adjustment of chemical reaction process; Step 1) For the industrial computer operating system of the processing object, call the number m of processing object samples stored in the database in the memory, the unit is one, and the processing evaluation index Q, the unit is mm/s; wherein the number m of processing object samples corresponds to the chemical reaction Materials and conditions include basic yellow 40 solvent concentration C1, unit is g/L, _TiO2 concentration C2, unit is g/L, ultraviolet light intensity I, unit is lx and temperature T, unit is ℃; processing evaluation index Q, The unit is mm/s, including the machining accuracy a in mm, the surface roughness r in μm and the processing time h in s. The larger the Q value, the better the processing quality. The mathematical formula is as follows:

Step 2) Establish the linear regression model function formula Q' as follows, w ₀ , w ₁ , w ₂ , and w ₃ are randomly initialized weight parameters: Q'＝w ₀ +w ₁ ×C1+w ₂ ×C2+w ₃ ×L+w ₄ ×T (2) Step 3) set up feature vector x, x is by basic yellow 40 solvent concentration C1, TiO Concentration C2, ultraviolet light intensity I, the vector that temperature T forms, x=[C1, C2, I, T]; Set up loss function, The loss value L ₁ is used to measure the difference between the linear regression model output Q' obtained by formula (2) and the actual processing evaluation index Q obtained by formula (1). The resulting formula is as follows, Q ⁱ and Q' ⁱ are respectively The actual evaluation index value and the predicted evaluation index value of the i-th processing:

Step 4) Use the stochastic batch gradient descent algorithm to iteratively update the weight value until the specified number of iterations is reached, preset 1000 times, calculate the weight parameter value when the loss value is the smallest, and use the following iteration formula to update the weight value, α is the learning rate Generally take 0.001, t represents the iteration round, t is set to 0 at the beginning of the iteration, and the random number initialization of the Gaussian distribution is generated by the program in advance,

A total of 1000 updates means that formula (4) needs to be calculated 1000 times, and each time you need to calculate the result from the previous calculation />

The four weight parameters are substituted into formula (4) to calculate this round of updated />

in

Both represent iterative calculations; Step 5) After the iteration is complete

That is, the corresponding weight values of w ₀ , w ₁ , w ₂ , and w ₃ can be substituted into the formula (2) to establish a prediction model. Through the prediction model, it is possible to predict the concentration of different basic yellow 40 solvents, the concentration of TiO2, and the intensity of ultraviolet light. The value of the processing evaluation index at the temperature; Step 6) When the Q' value is the largest according to the prediction model, that is, when the processing evaluation index is the highest, the values corresponding to the concentration of Basic Yellow 40 solvent C1, the concentration of TiO2, the intensity of ultraviolet light L, and the temperature T are obtained respectively; The system commands the parameters and communicates them to the UV lamp and the polishing liquid adding device, so as to achieve the purpose of optimizing and adjusting the chemical reaction process; (2) Optimizing and adjusting the polishing path of the robotic arm; Step 1. For the processing object, the operating system of the industrial computer calls the expected trajectory y _k stored in the cloud server database, which is composed of k coordinate points. The actual trajectory during the strategy network training process is y' _k , and the error function e _k is established. The formula is as follows : e _k ＝y _k -y' _k (5) Through the training of the strategy network model, the value of e _k is made as small as possible, that is, the error of the polishing trajectory is reduced; Step ② Set the reward value R according to the error value. The smaller the error is, the larger the reward value is. In order to allow the policy network to better understand the future and potential rewards, the future attenuated rewards are used here, assuming that the action is the second, and R _k is the second The reward obtained K times, its formula is as follows: R＝R ₁ +γR ₂ +γ ² R ₃ +...+γ ^k-1 R _k (γ is the attenuation coefficient of 0.99) (6); Step ③ Construct the neural network model, the input is the current coordinate value of the manipulator, and the output is the probability value matrix p=[p ₁ ,p ₂ ,p ₃ ,...,p _n ](n determined by the number of orientations set), which can be used to select actions; Step ④ Establish the loss function L ₂ , using the cross-entropy loss function, the formula is as follows: L ₂ =-(r*logp)/n (7) Increases the probability of actions with higher reward values and decreases the probability of actions with lower reward values; In step ⑤, the loss function is substituted, and the weight value in the neural network is iteratively updated using the gradient descent algorithm, and finally a policy network that can provide the optimal policy is obtained. Based on the policy network, the polishing path of the manipulator can be optimized and adjusted. 2. The processing platform based on photocatalytic cobalt-based alloy processing method construction as claimed in claim 1, characterized in that: comprising The robot module includes an industrial robot and a flexible pneumatic polishing device. The flexible polishing device is installed on the mechanical arm of the industrial robot to adjust the position of the workpiece; The polishing workbench includes a polishing disc for holding the polishing liquid and an illuminating device for irradiating the polishing liquid and the workpiece, the illuminating device is installed in the polishing disc, and the illuminating head of the illuminating device is aligned with the polishing liquid; The polishing liquid adding device is arranged next to the polishing workbench, and its liquid filling port is located directly above the polishing pool of the polishing workbench, and is used for adding polishing liquid to the polishing workbench; The industrial computer is electrically or signally connected to the control terminal of the robot module, polishing workbench and polishing liquid adding device, and is used to control the work of the robot module, polishing workbench and polishing liquid adding device, to realize the optimal adjustment of the chemical reaction process and the adjustment of the mechanical arm Polishing path optimization adjustment; And a memory, the signal transmission end of which is bidirectionally signal-connected or electrically connected to the signal transmission port of the industrial computer through a wire or a wireless network for storing processing information. 3. The processing platform according to claim 2, characterized in that: the bottom of the polishing disc is fixedly connected to a rotary table driven by a rotating motor to realize the rotation of the polishing disc around its own vertical central axis; An elastic layer for supporting abrasive grains, an elastic film layer, and an abrasive grain layer for processing the surface of the mold are provided in sequence, wherein the bottom of the elastic layer is connected to the inner bottom plate of the polishing pool, and the upper surface of the elastic layer is laid with an elastic film layer, and The edge of the elastic film layer is sealed and connected with the inner wall of the polishing pool, so that the elastic film layer and the inner wall of the polishing pool form a processing area for containing the polishing liquid; the abrasive grain layer is laid on the surface of the elastic film layer, and the abrasive grain layer is filled for Polishing fluid for lubricating and cooling; the control terminal of the rotating motor is electrically connected with the corresponding control terminal of the master controller. 4. The processing platform according to claim 3, characterized in that: the elastic layer is an array of conical coil springs, and the array of conical coil springs includes several mutually independent conical coil springs, and the conical coil springs The small diameter end of the said polishing pool is affixed to the inner bottom surface of said polishing pool, and the large diameter end of said conical coil spring is connected with a hemispherical support, and the arc curve of said hemispherical support is in contact with the lower surface of the elastic layer; The upper end of the conical coil spring is against the lower surface of the elastic film layer. 5. The processing platform according to claim 3, wherein the illumination device is an ultraviolet lamp fixed at the center of the rotary table, wherein the ultraviolet lamp is always irradiated on the workpiece processing area.

Images