Manufacturing method of pressure reducing valve, finishing parameter determination method and related equipment

Abstract

The application provides a manufacturing method of a pressure reducing valve, comprising the following steps: rough machining a first pressure reducing valve pipe and other assembly parts; in an environment without external light, a variable parallel light source is used to irradiate the inlet of the first pressure reducing valve pipe for a preset time, and an optical acquisition element is used to acquire the emergent light of the variable parallel light source passing through the first pressure reducing valve pipe at the outlet of the first pressure reducing valve pipe to obtain a first optical image sequence; based on the first optical image sequence, a fine machining parameter of the first pressure reducing valve pipe is matched; the first pressure reducing valve pipe is fine machined according to the fine machining parameter to obtain a second pressure reducing valve pipe after fine machining; and the second pressure reducing valve pipe is sent to an assembly process to be assembled with other assembly parts to obtain an assembled pressure reducing valve. The fine machining parameter in the fine machining stage is matched through the optical image sequence, so that the pressure reducing valve pipe can be more accurately fine machined in the fine machining stage, and the yield of the high-precision pressure reducing valve is improved.

Description

Technical Field

[0001] This invention relates to the field of precision machining, and in particular to a method for manufacturing a pressure reducing valve, a method for determining precision machining parameters, and related equipment. Background Technology

[0002] Traditional pressure reducing valve processing often involves setting the position and processing parameters before machining the internal flow channels of the valve fittings in one go. However, due to the inherent errors of the processing equipment, the precision of the processed pressure reducing valves is inconsistent. For pressure reducing valves with high precision requirements, traditional processing methods may not be accurate enough, resulting in a low assembly success rate and consequently a low yield. Summary of the Invention

[0003] This invention provides a method for manufacturing a pressure-reducing valve, aiming to solve the problem of insufficient precision in the existing high-precision pressure-reducing valve manufacturing process, resulting in low assembly success rate and consequently low yield. By performing rough machining on the pressure-reducing valve fittings while retaining a certain machining allowance, and then matching the finishing parameters in the finishing stage with optical image sequences, the finishing stage can perform more precise finishing on the pressure-reducing valve fittings. This significantly improves the machining accuracy and consistency of the pressure-reducing valve fittings, ensuring the stable and reliable performance of the final pressure-reducing valve fittings, thereby increasing the yield of high-precision pressure-reducing valves.

[0004] In a first aspect, embodiments of the present invention provide a method for manufacturing a pressure-reducing valve, the method comprising:

[0005] The first pressure-reducing valve fitting and other assemblies are roughly machined. The first pressure-reducing valve fitting has multi-stage pre-machined flow channels with different levels of machining allowance.

[0006] In an environment without external light, a variable parallel light source is used to illuminate the inlet of the first pressure-reducing valve fitting for a preset time. The light emitted from the variable parallel light source through the first pressure-reducing valve fitting is then collected at the outlet of the first pressure-reducing valve fitting by an optical acquisition element to obtain a first optical image sequence.

[0007] Based on the first optical image sequence, the finishing parameters of the first pressure reducing valve fitting are matched, and different first optical image sequences correspond to different finishing parameters;

[0008] The first pressure-reducing valve fitting is precision-machined using the aforementioned precision machining parameters to obtain the precision-machined second pressure-reducing valve fitting;

[0009] The second pressure-reducing valve fitting is sent into the assembly process to be assembled with other fittings to obtain the assembled pressure-reducing valve.

[0010] Optionally, matching the finishing parameters of the first pressure-reducing valve fitting based on the optical image sequence includes:

[0011] The first feature extraction network is used to extract the first spatial features of each frame of the first optical image in the first optical image sequence to obtain the first spatial feature sequence of the first optical image sequence. Each frame of the first optical image corresponds to a first spatial feature in the first spatial feature sequence.

[0012] The first spatial feature sequence is extracted by a first temporal network to obtain the first spatiotemporal features of the optical image sequence.

[0013] The first spatiotemporal features are processed by linear regression using a first linear regression network to obtain the finishing parameters of the first pressure-reducing valve fitting.

[0014] Optionally, before matching the finishing parameters of the first pressure-reducing valve fitting based on the first optical image sequence, the method further includes:

[0015] The machining allowance distribution of the sample pressure-reducing valve fitting is obtained by scanning the rough-machined sample pressure-reducing valve fitting.

[0016] The true values ​​of the finishing parameters of the sample pressure-reducing valve fitting are obtained based on the machining allowance distribution.

[0017] In an environment without external light, a variable parallel light source is used to illuminate the sample pressure-reducing valve fitting from the inlet for a preset time. The light emitted from the variable parallel light source through the sample pressure-reducing valve fitting is then collected by an optical acquisition element at the outlet of the sample pressure-reducing valve fitting to obtain a first sample optical image sequence.

[0018] The first sample optical image sequence is associated with the true value of the finishing parameters to obtain first sample data. Each first sample data includes a first sample optical image sequence and the corresponding true value of the finishing parameters.

[0019] The first feature extraction network, the first temporal network, and the first linear regression network to be trained are sequentially combined to obtain the first model to be trained.

[0020] The first model to be trained is trained in a supervised manner using the first sample data. Once training is complete, the first feature extraction network, the first temporal network, and the first linear regression network are obtained.

[0021] Secondly, embodiments of the present invention also provide a method for determining finishing parameters, applied to determining the finishing parameters of a first pressure-reducing valve fitting after rough machining. The first pressure-reducing valve fitting has multi-stage pre-machined flow channels, and the multi-stage pre-machined flow channels have different levels of machining allowance. The method for determining the finishing parameters of the pressure-reducing valve includes the following steps:

[0022] In an environment without external light, a variable parallel light source is used to illuminate the inlet of the first pressure-reducing valve fitting for a preset time. The light emitted from the variable parallel light source through the first pressure-reducing valve fitting is then collected at the outlet of the first pressure-reducing valve fitting by an optical acquisition element to obtain a first optical image sequence.

[0023] Based on the first optical image sequence, the finishing parameters of the first pressure reducing valve fitting are matched, and different first optical image sequences correspond to different finishing parameters.

[0024] Optionally, matching the finishing parameters of the first pressure-reducing valve fitting based on the optical image sequence includes:

[0025] The first feature extraction network is used to extract the first spatial features of each frame of the first optical image in the first optical image sequence to obtain the first spatial feature sequence of the first optical image sequence. Each frame of the first optical image corresponds to a first spatial feature in the first spatial feature sequence.

[0026] The first spatial feature sequence is extracted by a first temporal network to obtain the first spatiotemporal features of the optical image sequence.

[0027] The first spatiotemporal features are processed by linear regression using a first linear regression network to obtain the finishing parameters of the first pressure-reducing valve fitting.

[0028] Optionally, before matching the finishing parameters of the first pressure-reducing valve fitting based on the first optical image sequence, the method further includes:

[0029] The machining allowance distribution of the sample pressure-reducing valve fitting is obtained by scanning the rough-machined sample pressure-reducing valve fitting.

[0030] The true values ​​of the finishing parameters of the sample pressure-reducing valve fitting are obtained based on the machining allowance distribution.

[0031] In an environment without external light, a variable parallel light source is used to illuminate the sample pressure-reducing valve fitting from the inlet for a preset time. The light emitted from the variable parallel light source through the sample pressure-reducing valve fitting is then collected by an optical acquisition element at the outlet of the sample pressure-reducing valve fitting to obtain a first sample optical image sequence.

[0032] The first sample optical image sequence is associated with the true value of the finishing parameters to obtain first sample data. Each first sample data includes a first sample optical image sequence and the corresponding true value of the finishing parameters.

[0033] The first feature extraction network, the first temporal network, and the first linear regression network to be trained are sequentially combined to obtain the first model to be trained.

[0034] The first model to be trained is trained in a supervised manner using the first sample data. Once training is complete, the first feature extraction network, the first temporal network, and the first linear regression network are obtained.

[0035] Thirdly, embodiments of the present invention also provide a finishing parameter determination device, applied to determine the finishing parameters of a first pressure-reducing valve fitting after rough machining. The first pressure-reducing valve fitting has multi-stage pre-machined flow channels, the multi-stage pre-machined flow channels having different levels of machining allowance. The finishing parameter determination device for the pressure-reducing valve fitting includes:

[0036] The first acquisition module is used to illuminate the inlet of the first pressure-reducing valve fitting with a variable parallel light source for a preset time in an environment without external light, and to acquire the outgoing light of the variable parallel light source through the first pressure-reducing valve fitting through an optical acquisition element at the outlet of the first pressure-reducing valve fitting to obtain a first optical image sequence.

[0037] The processing module is used to match the finishing parameters of the first pressure reducing valve fitting based on the first optical image sequence, with different first optical image sequences corresponding to different finishing parameters.

[0038] Optionally, the processing module includes:

[0039] The first processing unit is configured to perform first spatial feature extraction on each frame of the first optical image in the first optical image sequence through a first feature extraction network to obtain a first spatial feature sequence of the first optical image sequence, wherein each frame of the first optical image corresponds to a first spatial feature in the first spatial feature sequence.

[0040] The second processing unit is used to perform first temporal feature extraction on the first spatial feature sequence through a first temporal network to obtain the first spatiotemporal features of the optical image sequence.

[0041] The third processing unit is used to perform linear regression processing on the first spatiotemporal features through the first linear regression network to obtain the finishing parameters of the first pressure-reducing valve fitting.

[0042] Optionally, the device further includes:

[0043] The scanning module is used to obtain the machining allowance distribution of the sample pressure-reducing valve fitting by scanning the rough-machined sample pressure-reducing valve fitting;

[0044] The acquisition module is used to obtain the true values ​​of the finishing parameters of the sample pressure-reducing valve fitting based on the machining allowance distribution;

[0045] The second acquisition module is used to illuminate the sample pressure-reducing valve fitting with a variable parallel light source for a preset time in an environment without external light, and to acquire the emitted light from the variable parallel light source through the sample pressure-reducing valve fitting through an optical acquisition element at the outlet of the sample pressure-reducing valve fitting to obtain a first sample optical image sequence.

[0046] The association module is used to associate the first sample optical image sequence with the true value of the finishing parameters to obtain first sample data. Each first sample data includes a first sample optical image sequence and the corresponding true value of the finishing parameters.

[0047] The synergy module is used to sequentially synergize the first feature extraction network to be trained, the first temporal network to be trained, and the first linear regression network to be trained to obtain the first model to be trained.

[0048] The training module is used to perform supervised training on the first model to be trained using the first sample data. After training is completed, the first feature extraction network, the first temporal network, and the first linear regression network are obtained.

[0049] Fourthly, embodiments of the present invention provide a processing and manufacturing system for a pressure reducing valve, the system comprising a roughing device, a finishing device, an assembly line, and a finishing parameter determining device as described in any of the embodiments of the present invention;

[0050] The roughing device is used to rough machine the first pressure-reducing valve fitting and other assemblies. The first pressure-reducing valve fitting has multi-stage pre-processing channels, and the multi-stage pre-processing channels have different levels of machining allowance.

[0051] The finishing device is used to finish the first pressure-reducing valve fitting using the finishing parameters to obtain the finished second pressure-reducing valve fitting;

[0052] The assembly line is used to send the second pressure-reducing valve fitting into the assembly process to assemble it with other fittings to obtain an assembled pressure-reducing valve.

[0053] In this embodiment of the invention, a first pressure-reducing valve fitting and other assemblies are roughly machined. The first pressure-reducing valve fitting has multi-stage pre-processed flow channels with different levels of machining allowance. In an environment without external light, a variable parallel light source is used to illuminate the inlet of the first pressure-reducing valve fitting for a preset time. At the outlet of the first pressure-reducing valve fitting, an optical acquisition element collects the emitted light from the variable parallel light source passing through the first pressure-reducing valve fitting, obtaining a first optical image sequence. Based on the first optical image sequence, the finishing parameters of the first pressure-reducing valve fitting are matched, with different first optical image sequences corresponding to different finishing parameters. The first pressure-reducing valve fitting is then finished using the finishing parameters to obtain a finished second pressure-reducing valve fitting. The second pressure-reducing valve fitting is then sent to the assembly process to be assembled with other assemblies to obtain an assembled pressure-reducing valve. By performing rough machining on the pressure reducing valve fittings while retaining a certain machining allowance, and matching the finishing parameters of the finishing stage with optical image sequences, the finishing stage can perform more precise finishing on the pressure reducing valve fittings. This significantly improves the machining accuracy and consistency of the pressure reducing valve fittings, ensuring the stable and reliable performance of the final pressure reducing valve fittings, thereby increasing the yield of high-precision pressure reducing valves. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart of a method for processing and manufacturing a pressure-reducing valve according to an embodiment of the present invention;

[0056] Figure 2 This is a flowchart of a method for determining finishing parameters provided in an embodiment of the present invention. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] like Figure 1 As shown, Figure 1 This is a flowchart illustrating a method for manufacturing a pressure-reducing valve according to an embodiment of the present invention. The method for manufacturing the pressure-reducing valve includes:

[0059] S1. Roughly machine the first pressure-reducing valve fitting and other assemblies.

[0060] In this embodiment of the invention, the first pressure-reducing valve fitting has multi-stage pre-processed flow channels, and the multi-stage pre-processed flow channels have different levels of machining allowance; other fittings may be valve cores, valve body shells, etc.

[0061] The purpose of the machining allowance is to prevent over-machining, which could lead to irreparable errors in the pre-machined flow channel structure. Maintaining a certain machining allowance during the roughing stage and then precisely removing it during the finishing stage reduces the probability of irreparable errors in the flow channel structure. The flow channel is the internal passage of the pressure-reducing valve, and pressure reduction control of the fluid is achieved through the flow channel and valve core. Different stages of the flow channel correspond to different inner diameters.

[0062] S2. In an environment without external light, a variable parallel light source is used to illuminate the inlet of the first pressure-reducing valve fitting for a preset time. At the outlet of the first pressure-reducing valve fitting, the light emitted by the variable parallel light source through the first pressure-reducing valve fitting is collected by an optical acquisition element to obtain a first optical image sequence.

[0063] In this embodiment of the invention, the environment without external light can be a black box environment. A variable parallel light source refers to a parallel light source with a variable wavelength. A parallel light source is a light source whose beam does not diverge or has a very small divergence angle, such as incident light or laser light. The optical acquisition element can be a CMOS image sensor or a PMT photomultiplier tube.

[0064] By illuminating the first pressure-reducing valve fitting with a variable parallel light source for a preset time at its inlet, and then collecting the emitted light from the first pressure-reducing valve fitting at its outlet, the influence of the pre-processed flow channel in the first pressure-reducing valve fitting on the light beam is considered. This influence is reflected in the first optical image sequence. This influence varies with different wavelengths of light beams; therefore, this dynamic influence can be expressed through the first optical image sequence. This dynamic influence is caused by the machining allowance. If the accuracy of the roughing device is consistent and there is no error in the roughing process, then the first optical image sequence for each first pressure-reducing valve fitting should be the same. If the first optical image sequences corresponding to different first pressure-reducing valve fittings are different, it indicates that the accuracy of the roughing device is inconsistent, and there is an error in the roughing process.

[0065] For the variation of the variable parallel light source, the wavelength should be kept strictly varied within a preset time to ensure that each pressure-reducing valve fitting can be irradiated with the same variable parallel light source.

[0066] S3. Based on the first optical image sequence, match the finishing parameters of the first pressure-reducing valve fitting.

[0067] In this embodiment of the invention, after obtaining the first optical image, the finishing parameters of the first pressure reducing valve fitting can be matched based on machine learning or optical analysis, and different first optical image sequences correspond to different finishing parameters.

[0068] S4. The first pressure-reducing valve fitting is precision machined using the precision machining parameters to obtain the precision-machined second pressure-reducing valve fitting.

[0069] In this embodiment of the invention, the first pressure-reducing valve fitting can be processed by a finishing device. The finishing device has a higher machining accuracy than the roughing device. The finishing device uses the finishing parameters obtained in step S3 to finish the first pressure-reducing valve fitting, thereby obtaining the finished second pressure-reducing valve fitting.

[0070] S5. The second pressure-reducing valve fitting is sent into the assembly process to be assembled with other fittings to obtain the assembled pressure-reducing valve.

[0071] In this embodiment of the invention, after obtaining the second pressure-reducing valve fitting, the second pressure-reducing valve fitting is sent to the assembly process and assembled with other fittings in the assembly process. The assembly process can be manual assembly or automatic assembly on the assembly line.

[0072] In this embodiment of the invention, a first pressure-reducing valve fitting and other assemblies are roughly machined. The first pressure-reducing valve fitting has multi-stage pre-processed flow channels with different levels of machining allowance. In an environment without external light, a variable parallel light source is used to illuminate the inlet of the first pressure-reducing valve fitting for a preset time. At the outlet of the first pressure-reducing valve fitting, an optical acquisition element collects the emitted light from the variable parallel light source passing through the first pressure-reducing valve fitting, obtaining a first optical image sequence. Based on the first optical image sequence, the finishing parameters of the first pressure-reducing valve fitting are matched, with different first optical image sequences corresponding to different finishing parameters. The first pressure-reducing valve fitting is then finished using the finishing parameters to obtain a finished second pressure-reducing valve fitting. The second pressure-reducing valve fitting is then sent to the assembly process to be assembled with other assemblies to obtain an assembled pressure-reducing valve. By performing rough machining on the pressure reducing valve fittings while retaining a certain machining allowance, and matching the finishing parameters of the finishing stage with optical image sequences, the finishing stage can perform more precise finishing on the pressure reducing valve fittings. This significantly improves the machining accuracy and consistency of the pressure reducing valve fittings, ensuring the stable and reliable performance of the final pressure reducing valve fittings, thereby increasing the yield of high-precision pressure reducing valves.

[0073] Optionally, the step of matching the finishing parameters of the first pressure-reducing valve fitting based on the optical image sequence includes: extracting first spatial features from each frame of the first optical image in the first optical image sequence using a first feature extraction network to obtain a first spatial feature sequence of the first optical image sequence, where each frame of the first optical image corresponds to a first spatial feature in the first spatial feature sequence; extracting first temporal features from the first spatial feature sequence using a first temporal network to obtain a first spatiotemporal feature of the optical image sequence; and performing linear regression processing on the first spatiotemporal feature using a first linear regression network to obtain the finishing parameters of the first pressure-reducing valve fitting.

[0074] In this embodiment of the invention, the first feature extraction network may be a convolutional neural network. The convolutional kernel of the convolutional neural network performs a convolution operation on each frame of the first optical image in the first optical image sequence to obtain the first optical feature corresponding to each frame of the first optical image. The optical feature may be a color feature and a position feature.

[0075] The first temporal network can be a recurrent neural network (RNN) or a long short-term memory network (LSMT). The first temporal network is used to extract temporal features of the first optical features in the time dimension to obtain the first spatiotemporal features of the optical image sequence.

[0076] The first linear regression network is a fully connected network. This fully connected network processes the first spatiotemporal features and outputs the finishing parameters for the first pressure-reducing valve fitting. The higher the dimensionality of the first spatiotemporal features, the more neurons the fully connected network has, and the more accurate the finishing parameters for the first pressure-reducing valve fitting will be.

[0077] In this embodiment of the invention, when the variable parallel light source passes through the first pressure-reducing valve fitting, it is affected by the wall of the pre-processed flow channel, causing a significant change in the light emitted from the first pressure-reducing valve fitting. This change is implicitly related to the processing allowance of the pre-processed flow channel, and this implicit relationship is contained in the first optical image acquired by the optical acquisition element. This relationship changes with the wavelength of the parallel light source. The variable parallel light source generates a parallel light source with a preset sequence of wavelengths over time, causing the light emitted from the first pressure-reducing valve fitting to change with the wavelength. This change is contained in the first optical image sequence acquired by the optical acquisition element within a preset time. This change corresponds to different processing allowances, and different processing allowances correspond to different finishing parameters.

[0078] Compared to traditional fluid measurement or scanning measurement, optical imaging combined with machine learning methods can more quickly determine the errors in the roughing stage, thereby matching more accurate finishing parameters to perform finishing on the first pressure reducing valve fitting, resulting in a high-precision pressure reducing valve.

[0079] Optionally, before matching the finishing parameters of the first pressure-reducing valve fitting based on the first optical image sequence, the method further includes: obtaining the machining allowance distribution of the sample pressure-reducing valve fitting by scanning a rough-machined sample pressure-reducing valve fitting; obtaining the true values ​​of the finishing parameters of the sample pressure-reducing valve fitting based on the machining allowance distribution; in an environment without external light, illuminating the sample pressure-reducing valve fitting from the inlet for a preset time using a variable parallel light source, and collecting the outgoing light from the sample pressure-reducing valve fitting through the variable parallel light source using an optical acquisition element at the outlet of the sample pressure-reducing valve fitting to obtain a first sample optical image sequence; associating the first sample optical image sequence with the true values ​​of the finishing parameters to obtain first sample data, each first sample data including a first sample optical image sequence and the corresponding true value of the finishing parameters; sequentially connecting the first feature extraction network to be trained, the first temporal network to be trained, and the first linear regression network to be trained to obtain a first training model; performing supervised training on the first training model using the first sample data, and obtaining the trained first feature extraction network, the first temporal network, and the first linear regression network.

[0080] In this embodiment of the invention, the first feature extraction network to be trained can be a convolutional neural network, the first temporal network to be trained can be a recurrent neural network (RNN) or a long short-term memory (LSMT) network, and the first linear regression network to be trained is a fully connected network. When performing simultaneous modeling, the output of the first feature extraction network to be trained is used as the input of the first temporal network to be trained, and the output of the first temporal network to be trained is used as the input of the first linear regression network to be trained, thereby obtaining the first model to be trained.

[0081] Sample pressure-reducing valve fittings with different machining allowances can be scanned to obtain the true machining allowance distribution of the sample pressure-reducing valve fittings with different machining allowances. The aforementioned true machining allowance distribution refers to the true machining allowance of each stage of pre-processed flow channel. Based on the true machining allowance distribution, the true finishing parameters of the sample pressure-reducing valve fittings can be determined, and thus the true values ​​of the finishing parameters can be obtained. The true values ​​of the finishing parameters can be used as labels to guide the first model to be trained.

[0082] For each sample pressure-reducing valve fitting, a first sample image sequence is acquired. For changes in the variable parallel light source, the wavelength should be strictly varied within a preset time to ensure that each sample pressure-reducing valve fitting is illuminated by the same variable parallel light source. The first sample image sequence of each sample pressure-reducing valve fitting is associated with its corresponding ground truth value of the finishing parameters. This ground truth value of the finishing parameters serves as a label for the first sample image sequence, guiding the first training model to output predicted finishing parameters that are the same as or similar to the ground truth value of the finishing parameters after inputting the first sample image sequence.

[0083] The first training model is trained in a supervised manner using the first sample data. The training is completed when the prediction count is reached or the first training model converges, and the trained first feature extraction network, first temporal network, and first linear regression network are obtained.

[0084] Optionally, the accuracy of the second pressure-reducing valve fitting or the assembled pressure-reducing valve can be tested. The accuracy test can be performed in an environment without external light by irradiating the inlet of the second pressure-reducing valve fitting or the assembled pressure-reducing valve with a variable parallel light source for a preset time, and then collecting the emitted light from the variable parallel light source through the second pressure-reducing valve fitting or the assembled pressure-reducing valve with an optical acquisition element at the outlet of the second pressure-reducing valve fitting or the assembled pressure-reducing valve to obtain a second optical image sequence.

[0085] Based on the second optical image sequence, the accuracy of the second pressure-reducing valve fitting or the assembled pressure-reducing valve is determined. When the accuracy of the second pressure-reducing valve fitting meets the preset accuracy requirements, the second pressure-reducing valve fitting is sent to the assembly process to be assembled with other components to obtain the assembled pressure-reducing valve. Alternatively, when the accuracy of the assembled pressure-reducing valve meets the preset accuracy requirements, the assembled pressure-reducing valve can be determined to be a qualified high-precision pressure-reducing valve.

[0086] Specifically, a second spatial feature extraction network can be used to extract the second spatial features of each frame of the second optical image sequence to obtain the second spatial feature sequence of the second optical image sequence, with each frame of the second optical image corresponding to a second spatial feature in the second spatial feature sequence; a second temporal network can be used to extract the second temporal features of the second spatial feature sequence to obtain the second spatiotemporal features of the optical image sequence; and a second linear regression network can be used to perform linear regression processing on the second spatiotemporal features to obtain the accuracy of the second pressure reducing valve fitting or the assembled pressure reducing valve.

[0087] The true accuracy of the sample second pressure-reducing valve fitting or sample pressure-reducing valve can be determined by scanning the finely processed sample second pressure-reducing valve fitting or sample pressure-reducing valve. In an environment without external light, a variable parallel light source is used to illuminate the sample second pressure-reducing valve fitting or sample pressure-reducing valve from the inlet for a preset time. At the outlet of the sample second pressure-reducing valve fitting or sample pressure-reducing valve, the emitted light from the variable parallel light source passing through the sample second pressure-reducing valve fitting or sample pressure-reducing valve is collected by an optical acquisition element to obtain a second sample optical image sequence. The second sample optical image sequence is associated with the true accuracy to obtain second sample data. Each second sample data includes a second sample optical image sequence and the corresponding true accuracy. The second feature extraction network, the second temporal network, and the second linear regression network to be trained are sequentially connected to obtain the second training model. The second training model is then trained in a supervised manner using the second sample data. After training is completed, the trained second feature extraction network, second temporal network, and second linear regression network are obtained.

[0088] In this embodiment of the invention, the second pressure-reducing valve fitting or pressure-reducing valve is subjected to precision detection by a second optical image sequence, which can improve the detection speed and thus improve the precision detection efficiency of the high-precision pressure-reducing valve.

[0089] Secondly, such as Figure 2 This is a flowchart of a method for determining finishing parameters provided in an embodiment of the present invention. The method is applied to determine the finishing parameters of a first pressure-reducing valve fitting after rough machining. The first pressure-reducing valve fitting has multi-stage pre-machined flow channels with different levels of machining allowance. The method for determining the finishing parameters of the pressure-reducing valve includes the following steps:

[0090] 201. In an environment without external light, a variable parallel light source is used to illuminate the inlet of the first pressure-reducing valve fitting for a preset time. At the outlet of the first pressure-reducing valve fitting, the light emitted by the variable parallel light source through the first pressure-reducing valve fitting is collected by an optical acquisition element to obtain a first optical image sequence.

[0091] In this embodiment of the invention, the environment without external light can be a black box environment. A variable parallel light source refers to a parallel light source with a variable wavelength. A parallel light source is a light source whose beam does not diverge or has a very small divergence angle, such as incident light or laser light. The optical acquisition element can be a CMOS image sensor or a PMT photomultiplier tube.

[0092] By illuminating the first pressure-reducing valve fitting with a variable parallel light source for a preset time at its inlet, and then collecting the emitted light from the first pressure-reducing valve fitting at its outlet, the influence of the pre-processed flow channel in the first pressure-reducing valve fitting on the light beam is considered. This influence is reflected in the first optical image sequence. This influence varies with different wavelengths of light beams; therefore, this dynamic influence can be expressed through the first optical image sequence. This dynamic influence is caused by the machining allowance. If the accuracy of the roughing device is consistent and there is no error in the roughing process, then the first optical image sequence for each first pressure-reducing valve fitting should be the same. If the first optical image sequences corresponding to different first pressure-reducing valve fittings are different, it indicates that the accuracy of the roughing device is inconsistent, and there is an error in the roughing process.

[0093] For the variation of the variable parallel light source, the wavelength should be kept strictly varied within a preset time to ensure that each pressure-reducing valve fitting can be irradiated with the same variable parallel light source.

[0094] 202. Based on the first optical image sequence, the finishing parameters of the first pressure reducing valve fitting are matched, and different first optical image sequences correspond to different finishing parameters.

[0095] In this embodiment of the invention, after obtaining the first optical image, the finishing parameters of the first pressure reducing valve fitting can be matched based on machine learning or optical analysis, and different first optical image sequences correspond to different finishing parameters.

[0096] In this embodiment of the invention, by performing rough machining on the pressure reducing valve fittings while retaining a certain machining allowance, and by matching the finishing parameters of the finishing stage with optical image sequences, the finishing stage can perform more precise finishing on the pressure reducing valve fittings. This can significantly improve the machining accuracy and consistency of the pressure reducing valve fittings, ensure the stable and reliable performance of the final pressure reducing valve fittings, and thus improve the yield of high-precision pressure reducing valves.

[0097] Optionally, the step of matching the finishing parameters of the first pressure-reducing valve fitting based on the optical image sequence includes: extracting first spatial features from each frame of the first optical image in the first optical image sequence using a first feature extraction network to obtain a first spatial feature sequence of the first optical image sequence, where each frame of the first optical image corresponds to a first spatial feature in the first spatial feature sequence; extracting first temporal features from the first spatial feature sequence using a first temporal network to obtain a first spatiotemporal feature of the optical image sequence; and performing linear regression processing on the first spatiotemporal feature using a first linear regression network to obtain the finishing parameters of the first pressure-reducing valve fitting.

[0098] In this embodiment of the invention, the first feature extraction network may be a convolutional neural network. The convolutional kernel of the convolutional neural network performs a convolution operation on each frame of the first optical image in the first optical image sequence to obtain the first optical feature corresponding to each frame of the first optical image. The optical feature may be a color feature and a position feature.

[0099] The first temporal network can be a recurrent neural network (RNN) or a long short-term memory network (LSMT). The first temporal network is used to extract temporal features of the first optical features in the time dimension to obtain the first spatiotemporal features of the optical image sequence.

[0100] The first linear regression network is a fully connected network. This fully connected network processes the first spatiotemporal features and outputs the finishing parameters for the first pressure-reducing valve fitting. The higher the dimensionality of the first spatiotemporal features, the more neurons the fully connected network has, and the more accurate the finishing parameters for the first pressure-reducing valve fitting will be.

[0101] In this embodiment of the invention, when the variable parallel light source passes through the first pressure-reducing valve fitting, it is affected by the wall of the pre-processed flow channel, causing a significant change in the light emitted from the first pressure-reducing valve fitting. This change is implicitly related to the processing allowance of the pre-processed flow channel, and this implicit relationship is contained in the first optical image acquired by the optical acquisition element. This relationship changes with the wavelength of the parallel light source. The variable parallel light source generates a parallel light source with a preset sequence of wavelengths over time, causing the light emitted from the first pressure-reducing valve fitting to change with the wavelength. This change is contained in the first optical image sequence acquired by the optical acquisition element within a preset time. This change corresponds to different processing allowances, and different processing allowances correspond to different finishing parameters.

[0102] Compared to traditional fluid measurement or scanning measurement, optical imaging combined with machine learning methods can more quickly determine the errors in the roughing stage, thereby matching more accurate finishing parameters to perform finishing on the first pressure reducing valve fitting, resulting in a high-precision pressure reducing valve.

[0103] Optionally, before matching the finishing parameters of the first pressure-reducing valve fitting based on the first optical image sequence, the method further includes: obtaining the machining allowance distribution of the sample pressure-reducing valve fitting by scanning a rough-machined sample pressure-reducing valve fitting; obtaining the true values ​​of the finishing parameters of the sample pressure-reducing valve fitting based on the machining allowance distribution; in an environment without external light, illuminating the sample pressure-reducing valve fitting from the inlet for a preset time using a variable parallel light source, and collecting the outgoing light from the sample pressure-reducing valve fitting through the variable parallel light source using an optical acquisition element at the outlet of the sample pressure-reducing valve fitting to obtain a first sample optical image sequence; associating the first sample optical image sequence with the true values ​​of the finishing parameters to obtain first sample data, each first sample data including a first sample optical image sequence and the corresponding true value of the finishing parameters; sequentially connecting the first feature extraction network to be trained, the first temporal network to be trained, and the first linear regression network to be trained to obtain a first training model; performing supervised training on the first training model using the first sample data, and obtaining the trained first feature extraction network, the first temporal network, and the first linear regression network.

[0104] In this embodiment of the invention, the first feature extraction network to be trained can be a convolutional neural network, the first temporal network to be trained can be a recurrent neural network (RNN) or a long short-term memory (LSMT) network, and the first linear regression network to be trained is a fully connected network. When performing simultaneous modeling, the output of the first feature extraction network to be trained is used as the input of the first temporal network to be trained, and the output of the first temporal network to be trained is used as the input of the first linear regression network to be trained, thereby obtaining the first model to be trained.

[0105] Sample pressure-reducing valve fittings with different machining allowances can be scanned to obtain the true machining allowance distribution of the sample pressure-reducing valve fittings with different machining allowances. The aforementioned true machining allowance distribution refers to the true machining allowance of each stage of pre-processed flow channel. Based on the true machining allowance distribution, the true finishing parameters of the sample pressure-reducing valve fittings can be determined, and thus the true values ​​of the finishing parameters can be obtained. The true values ​​of the finishing parameters can be used as labels to guide the first model to be trained.

[0106] For each sample pressure-reducing valve fitting, a first sample image sequence is acquired. For changes in the variable parallel light source, the wavelength should be strictly varied within a preset time to ensure that each sample pressure-reducing valve fitting is illuminated by the same variable parallel light source. The first sample image sequence of each sample pressure-reducing valve fitting is associated with its corresponding ground truth value of the finishing parameters. This ground truth value of the finishing parameters serves as a label for the first sample image sequence, guiding the first training model to output predicted finishing parameters that are the same as or similar to the ground truth value of the finishing parameters after inputting the first sample image sequence.

[0107] The first training model is trained in a supervised manner using the first sample data. The training is completed when the prediction count is reached or the first training model converges, and the trained first feature extraction network, first temporal network, and first linear regression network are obtained.

[0108] Thirdly, embodiments of the present invention also provide a finishing parameter determination device, applied to determine the finishing parameters of a first pressure-reducing valve fitting after rough machining. The first pressure-reducing valve fitting has multi-stage pre-machined flow channels, the multi-stage pre-machined flow channels having different levels of machining allowance. The finishing parameter determination device for the pressure-reducing valve fitting includes:

[0109] The first acquisition module is used to illuminate the inlet of the first pressure-reducing valve fitting with a variable parallel light source for a preset time in an environment without external light, and to acquire the outgoing light of the variable parallel light source through the first pressure-reducing valve fitting through an optical acquisition element at the outlet of the first pressure-reducing valve fitting to obtain a first optical image sequence.

[0110] The processing module is used to match the finishing parameters of the first pressure reducing valve fitting based on the first optical image sequence, with different first optical image sequences corresponding to different finishing parameters.

[0111] Optionally, the processing module includes:

[0112] The first processing unit is configured to perform first spatial feature extraction on each frame of the first optical image in the first optical image sequence through a first feature extraction network to obtain a first spatial feature sequence of the first optical image sequence, wherein each frame of the first optical image corresponds to a first spatial feature in the first spatial feature sequence.

[0113] The second processing unit is used to perform first temporal feature extraction on the first spatial feature sequence through a first temporal network to obtain the first spatiotemporal features of the optical image sequence.

[0114] The third processing unit is used to perform linear regression processing on the first spatiotemporal features through the first linear regression network to obtain the finishing parameters of the first pressure-reducing valve fitting.

[0115] Optionally, the device further includes:

[0116] The scanning module is used to obtain the machining allowance distribution of the sample pressure-reducing valve fitting by scanning the rough-machined sample pressure-reducing valve fitting;

[0117] The acquisition module is used to obtain the true values ​​of the finishing parameters of the sample pressure-reducing valve fitting based on the machining allowance distribution;

[0118] The second acquisition module is used to illuminate the sample pressure-reducing valve fitting with a variable parallel light source for a preset time in an environment without external light, and to acquire the emitted light from the variable parallel light source through the sample pressure-reducing valve fitting through an optical acquisition element at the outlet of the sample pressure-reducing valve fitting to obtain a first sample optical image sequence.

[0119] The association module is used to associate the first sample optical image sequence with the true value of the finishing parameters to obtain first sample data. Each first sample data includes a first sample optical image sequence and the corresponding true value of the finishing parameters.

[0120] The synergy module is used to sequentially synergize the first feature extraction network to be trained, the first temporal network to be trained, and the first linear regression network to be trained to obtain the first model to be trained.

[0121] The training module is used to perform supervised training on the first model to be trained using the first sample data. After training is completed, the first feature extraction network, the first temporal network, and the first linear regression network are obtained.

[0122] Fourthly, embodiments of the present invention provide a processing and manufacturing system for a pressure reducing valve, the system comprising a roughing device, a finishing device, an assembly line, and a finishing parameter determining device as described in any of the embodiments of the present invention;

[0123] The roughing device is used to rough machine the first pressure-reducing valve fitting and other assemblies. The first pressure-reducing valve fitting has multi-stage pre-processing channels, and the multi-stage pre-processing channels have different levels of machining allowance.

[0124] The finishing device is used to finish the first pressure-reducing valve fitting using the finishing parameters to obtain the finished second pressure-reducing valve fitting;

[0125] The assembly line is used to send the second pressure-reducing valve fitting into the assembly process to assemble it with other fittings to obtain an assembled pressure-reducing valve.

[0126] The above description discloses only preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A method for manufacturing a pressure-reducing valve, characterized in that, The manufacturing method of the pressure reducing valve includes: The first pressure-reducing valve fitting and other assemblies are roughly machined. The first pressure-reducing valve fitting has multi-stage pre-machined flow channels with different levels of machining allowance. In an environment without external light, a variable parallel light source is used to illuminate the inlet of the first pressure-reducing valve fitting for a preset time. The emitted light from the variable parallel light source through the first pressure-reducing valve fitting is then collected at the outlet of the first pressure-reducing valve fitting by an optical acquisition element to obtain a first optical image sequence. The variable parallel light source refers to a parallel light source with a variable wavelength. Based on the first optical image sequence, the finishing parameters of the first pressure reducing valve fitting are matched, and different first optical image sequences correspond to different finishing parameters; The first pressure-reducing valve fitting is precision-machined using the aforementioned precision machining parameters to obtain the precision-machined second pressure-reducing valve fitting; The second pressure-reducing valve fitting is sent into the assembly process and assembled with other fittings to obtain the assembled pressure-reducing valve. The process of matching the finishing parameters of the first pressure-reducing valve fitting based on the optical image sequence includes: The first feature extraction network is used to extract the first spatial features of each frame of the first optical image in the first optical image sequence to obtain the first spatial feature sequence of the first optical image sequence. Each frame of the first optical image corresponds to a first spatial feature in the first spatial feature sequence. The first spatial feature sequence is extracted by a first temporal network to obtain the first spatiotemporal features of the optical image sequence. The first spatiotemporal features are linearly regressed using a first linear regression network to obtain the finishing parameters of the first pressure-reducing valve fitting. Before matching the finishing parameters of the first pressure-reducing valve fitting based on the first optical image sequence, the method further includes: The machining allowance distribution of the sample pressure-reducing valve fitting is obtained by scanning the rough-machined sample pressure-reducing valve fitting. The true values ​​of the finishing parameters of the sample pressure-reducing valve fitting are obtained based on the machining allowance distribution. In an environment without external light, a variable parallel light source is used to illuminate the sample pressure-reducing valve fitting from the inlet for a preset time. The light emitted from the variable parallel light source through the sample pressure-reducing valve fitting is then collected by an optical acquisition element at the outlet of the sample pressure-reducing valve fitting to obtain a first sample optical image sequence. The first sample optical image sequence is associated with the true value of the finishing parameters to obtain first sample data. Each first sample data includes a first sample optical image sequence and the corresponding true value of the finishing parameters. The first feature extraction network, the first temporal network, and the first linear regression network to be trained are sequentially combined to obtain the first model to be trained. The first model to be trained is trained in a supervised manner using the first sample data. Once training is complete, the first feature extraction network, the first temporal network, and the first linear regression network are obtained.

2. A method for determining finishing parameters, characterized in that, The method for determining the finishing parameters of a pressure-reducing valve fitting, which has multiple pre-machined flow channels with different levels of machining allowance, includes the following steps: In an environment without external light, a first optical image sequence is obtained by illuminating the inlet of the first pressure-reducing valve fitting with a variable parallel light source for a preset time, and then collecting the emitted light from the variable parallel light source through the first pressure-reducing valve fitting at the outlet of the first pressure-reducing valve fitting using an optical acquisition element. Based on the first optical image sequence, the finishing parameters of the first pressure reducing valve fitting are matched, and different first optical image sequences correspond to different finishing parameters; The process of matching the finishing parameters of the first pressure-reducing valve fitting based on the optical image sequence includes: The first feature extraction network is used to extract the first spatial features of each frame of the first optical image in the first optical image sequence to obtain the first spatial feature sequence of the first optical image sequence. Each frame of the first optical image corresponds to a first spatial feature in the first spatial feature sequence. The first spatial feature sequence is extracted by a first temporal network to obtain the first spatiotemporal features of the optical image sequence. The first spatiotemporal features are linearly regressed using a first linear regression network to obtain the finishing parameters of the first pressure-reducing valve fitting. Before matching the finishing parameters of the first pressure-reducing valve fitting based on the first optical image sequence, the method further includes: The machining allowance distribution of the sample pressure-reducing valve fitting is obtained by scanning the rough-machined sample pressure-reducing valve fitting. The true values ​​of the finishing parameters of the sample pressure-reducing valve fitting are obtained based on the machining allowance distribution. In an environment without external light, a variable parallel light source is used to illuminate the sample pressure-reducing valve fitting from the inlet for a preset time. The light emitted from the variable parallel light source through the sample pressure-reducing valve fitting is then collected by an optical acquisition element at the outlet of the sample pressure-reducing valve fitting to obtain a first sample optical image sequence. The first sample optical image sequence is associated with the true value of the finishing parameters to obtain first sample data. Each first sample data includes a first sample optical image sequence and the corresponding true value of the finishing parameters. The first feature extraction network, the first temporal network, and the first linear regression network to be trained are sequentially combined to obtain the first model to be trained. The first model to be trained is trained in a supervised manner using the first sample data. Once training is complete, the first feature extraction network, the first temporal network, and the first linear regression network are obtained.

3. A device for determining finishing parameters, characterized in that, The device for determining the finishing parameters of a first pressure-reducing valve fitting, which has multiple pre-machined flow channels with different levels of machining allowance, is used in rough machining. The first acquisition module is used to illuminate the inlet of the first pressure-reducing valve fitting with a variable parallel light source for a preset time in an environment without external light, and to acquire the outgoing light of the variable parallel light source through the first pressure-reducing valve fitting through an optical acquisition element at the outlet of the first pressure-reducing valve fitting to obtain a first optical image sequence. The processing module is used to match the finishing parameters of the first pressure reducing valve fitting based on the first optical image sequence, and different first optical image sequences correspond to different finishing parameters; The processing module includes: The first processing unit is configured to perform first spatial feature extraction on each frame of the first optical image in the first optical image sequence through a first feature extraction network to obtain a first spatial feature sequence of the first optical image sequence, wherein each frame of the first optical image corresponds to a first spatial feature in the first spatial feature sequence. The second processing unit is used to perform first temporal feature extraction on the first spatial feature sequence through a first temporal network to obtain the first spatiotemporal features of the optical image sequence. The third processing unit is used to perform linear regression processing on the first spatiotemporal features through the first linear regression network to obtain the finishing parameters of the first pressure-reducing valve fitting. The device further includes: The scanning module is used to obtain the machining allowance distribution of the sample pressure-reducing valve fitting by scanning the rough-machined sample pressure-reducing valve fitting; The acquisition module is used to obtain the true values ​​of the finishing parameters of the sample pressure-reducing valve fitting based on the machining allowance distribution; The second acquisition module is used to illuminate the sample pressure-reducing valve fitting with a variable parallel light source for a preset time in an environment without external light, and to acquire the emitted light from the variable parallel light source through the sample pressure-reducing valve fitting through an optical acquisition element at the outlet of the sample pressure-reducing valve fitting to obtain a first sample optical image sequence. The association module is used to associate the first sample optical image sequence with the true value of the finishing parameters to obtain first sample data. Each first sample data includes a first sample optical image sequence and the corresponding true value of the finishing parameters. The synergy module is used to sequentially synergize the first feature extraction network to be trained, the first temporal network to be trained, and the first linear regression network to be trained to obtain the first model to be trained. The training module is used to perform supervised training on the first model to be trained using the first sample data. After training is completed, the first feature extraction network, the first temporal network, and the first linear regression network are obtained.

Images