Assembly quality inspection and calibration method and system based on gyroscope assembly process

Abstract

The present invention relates to the technical field of gyroscope assembly, and in particular to an assembly quality detection calibration method and system based on the gyroscope assembly process, which includes data collection and input, construction of an LSTM neural network model, model prediction, model output, output of the model assembly quality prediction result and adjustment parameter coordinates, and adjustment of a mechanical arm assembly according to a control signal. The present invention collects data of assembly gap and electrode alignment in real time, and normalizes the collected data, uses the trained neural network model, predicts the model and outputs the prediction result, and then adjusts and assembles through a mechanical arm assembly; converts the predicted adjustment parameter coordinates into a control signal of the mechanical arm assembly, automatically adjusts the components in the three directions of x, y, and z, and completes the assembly accuracy adjustment of a hemispherical resonator and a readout base according to the prediction result and parameter coordinates output by the model, thereby meeting the calibration requirements.

Description

Technical Field

[0001] The present invention relates to the technical field of gyroscope assembly, and in particular to an assembly quality detection and calibration method and system in a gyroscope assembly process. Background Art

[0002] Hemispherical resonator gyroscope (HRG) is a high-precision solid-state gyroscope with inertial navigation performance. It has no high-speed rotor, no bearings and friction-related components, and has the advantages of simple structure, long life, high reliability and radiation resistance. It is widely used in aerospace, military and high-precision instruments. HRG is mainly composed of a hemispherical resonator, an electrostatic excitation cover and a sensitive readout base. The hemispherical resonator is the core sensitive component of the gyroscope, with a high quality factor Q value and a stable natural vibration frequency.

[0003] The hemispherical resonator, electrostatic excitation cover and readout base of the gyroscope are welded together and sealed in a high vacuum container to form an angle or angular velocity sensor. Several discrete electrodes are distributed on the electrostatic excitation cover, forming several capacitors with the spherical surface of the resonator. The static high voltage generates the force required for the resonator to vibrate through these capacitors, forming micro-vibration of the resonator. The assembly standard of the hemispherical resonator and the readout base affects the assembly quality of the gyroscope, among which the assembly gap and electrode alignment are two important indicators in the assembly.

[0004] However, in the prior art, the assembly process of the hemispherical resonator and the readout base involves the coordinated use of an assembly mechanism, a detection mechanism, and an adjustment mechanism. After the assembly is completed, the assembly gap and electrode alignment need to be detected by the detection mechanism, and then the adjustment mechanism is manually controlled according to the detection result parameters. Manual calibration is required between the mechanisms. This not only increases the labor cost, but also requires a certain amount of time to be repeatedly adjusted during the assembly process, which is not conducive to improving and optimizing the assembly efficiency. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides an assembly quality detection and calibration method and system during the gyroscope assembly process.

[0006] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] The assembly quality inspection and calibration method and system based on the gyroscope assembly process include the following steps:

[0008] S1: Data collection and input, collecting assembly gap and electrode alignment data from the assembly line, including historical assembly data and real-time monitoring data, and normalizing them;

[0009] S2: Build LSTM neural network model;

[0010] S3: Model prediction, inputting the real-time monitored assembly gap and electrode alignment data into the model to obtain the predicted adjustment parameter coordinates;

[0011] S4: Model output, outputting the model assembly quality prediction result and adjustment parameter coordinates, converting the predicted adjustment parameter coordinates into control signals of the robot arm assembly, wherein the adjustment parameter coordinates include components in three directions: x, y, and z;

[0012] S5: The robotic arm assembly adjusts the first fastening mechanism for clamping the hemispherical shell in the vertical vector z direction and adjusts the second fastening mechanism for clamping the readout base in the horizontal vector x, y direction according to the control signal;

[0013] Wherein, the step S2 specifically includes the following sub-steps:

[0014] S201: extracting normalized assembly gap features and electrode alignment features;

[0015] S202: Use the normalized assembly gap features and electrode alignment features to train the LSTM model, and the output of the model is the adjustment parameter coordinates.

[0016] Furthermore, in step S1, the assembly gap is , the electrode alignment is , the normalization formula is:

[0017] ; .

[0018] in, is the maximum value of the assembly clearance, is the minimum value of the assembly clearance, It is the normalized assembly gap value calculated by linearly mapping the actual assembly gap of the sample with the minimum and maximum values ​​in the entire sample;

[0019] is the maximum value of electrode alignment, is the minimum value of electrode alignment, It is the normalized electrode alignment value calculated by linearly mapping the actual electrode alignment of the sample with the minimum and maximum values ​​in the sample.

[0020] Furthermore, in step S202, the hidden layer of the LSTM model is used to process the time series characteristics of the assembly gap and the electrode alignment, and the expression formula is:

[0021]

[0022] in, is the time step The hidden state of is the normalized input feature, and is the weight matrix, is the bias vector.

[0023] Furthermore, in step S4, the features of the hidden layer of LSTM are mapped to the output features, and the formula is: ,in, is the final hidden state of the hidden layer, are the predicted adjustment parameter coordinates.

[0024] Furthermore, in step S4, by calculating the difference between the predicted value and the actual value of the assembly gap and electrode alignment of each sample, the squares of these differences are added and then divided by the number of samples to obtain the final mean square error, thereby improving the prediction accuracy. The formula is:

[0025]

[0026] in, is the sample size, and They are The predicted values ​​of assembly gap and electrode alignment for samples, and It is The actual values ​​of assembly gap and electrode alignment of the samples, and are the minimum and maximum values ​​of the assembly clearance, and are the minimum and maximum values ​​of electrode alignment.

[0027] Furthermore, in step S4, the model outputs the prediction result formula of the assembly quality as follows: ,in It is an LSTM neural network model.

[0028] Furthermore, in step S4, the coordinate formula of the adjustment parameter predicted by the model is calculated and output as: ,in, , , are the components of the adjustment parameters in the three directions.

[0029] An assembly quality inspection and calibration system based on the assembly quality inspection and calibration method during the assembly process of a gyroscope includes a mechanical arm component and a clamping component, wherein the clamping component is arranged at the control end of the mechanical arm component, and the clamping component includes a first fixed platform and a second fixed platform, wherein the first fixed platform is a columnar structure with a bottom opening and a fixed cavity, wherein the lower end of the fixed cavity is provided with a plurality of first fastening mechanisms for fixing a hemispherical shell, wherein the second fixed platform is sleeved on the lower end of the first fixed platform, wherein a first positioning plate is arranged around the bottom outer wall of the first fixed platform, wherein a second positioning plate is arranged around the top inner wall of the second fixed platform, wherein the first positioning plate and the second positioning plate are connected via a first elastic mechanism, wherein a plurality of second fastening mechanisms for adjusting a readout base are arranged on the inner wall of the second fixed platform, wherein the plurality of second fastening mechanisms are each equipped with a corresponding control motor, and wherein the control motor is connected to the control unit of the mechanical arm component.

[0030] First, the control end of the mechanical arm assembly rotates to the transport line of the hemispherical shell, so that the center line of the clamping assembly is aligned with the central axis of the hemispherical shell, and the clamping assembly is moved down. At this time, the first elastic mechanism is stretched until the first elastic mechanism reaches the maximum stretching state, and the bottom of the first fixed platform touches the placement plane of the hemispherical shell, and the hemispherical shell enters the fixed cavity. The hemispherical shell is limited and clamped by the first fastening mechanism. After the fixation is completed, the clamping assembly is moved up, the first elastic mechanism restores the elastic deformation, and the first fixed platform is moved up to form a receiving cavity with the second fixed platform that can accommodate the readout base. Then the control end of the mechanical part assembly rotates, and the clamping assembly fixed with the hemispherical shell is moved to the assembly area, where the readout base is placed. The clamping assembly moves down, aligns and assembles the hemispherical shell with the readout base, and controls each control motor respectively through the components of the three directions of the adjustment parameter coordinates x, y, and z calculated, so that the readout base is adjusted in the horizontal direction, and the y-axis direction is controlled by the up and down movement of the clamping assembly, and finally the function of precise adjustment is achieved.

[0031] Furthermore, the first fastening mechanism includes a driving rod and a horizontal push rod, the side wall of the first fixed platform is provided with several vertical cavities for the driving rod to move, the bottom surface of the driving rod is inclined, one end of the horizontal push rod is provided with an inclined surface matching the bottom surface of the driving rod, the other end of the horizontal push rod passes through the vertical cavity and the fixed cavity, a limiting ring and a second elastic mechanism are provided on the horizontal push rod, and the driving rod will push the horizontal push rod toward the fixed cavity when it moves downward, a pressure plate is provided above the first fixed platform, the tops of several driving rods pass through the top surface of the first fixed platform and are connected to the pressure plate, and an output mechanism is connected above the pressure plate through an adjustment rod passing through the mechanical arm assembly.

[0032] A longitudinally movable output mechanism is arranged in the control end of the mechanical arm assembly, and the lifting and lowering of the adjustment rod is controlled by the output mechanism, thereby controlling the lifting and lowering of several driving rods. A limiting portion cooperating with a limiting ring is arranged on the movable cavity of the horizontal push rod, and the second elastic mechanism is located between the limiting ring and the limiting portion. When the driving rod descends, the inclined surface at the bottom of the driving rod pushes the horizontal push rod to move into the fixed cavity, thereby achieving the function of clamping and fixing the hemispherical shell. After completing the assembly and welding of the hemispherical shell and the readout base, the control end of the mechanical arm assembly rotates to the finished product placement area, and by moving the driving rod upward, the second elastic mechanism restores the elastic deformation, and the limiting ring moves away from the limiting portion, so that the horizontal push rod rebounds into the vertical cavity, loosening the fixation of the hemispherical shell.

[0033] Furthermore, a plurality of the second fastening mechanisms are arranged at equal intervals, at least four of the second fastening mechanisms are provided, and the second fastening mechanisms have the same structure as the first fastening mechanisms. The second fastening mechanisms are controlled by corresponding control motors to achieve independent adjustment. Each control motor receives different control signals according to the unified regulation of the control signal, and achieves different movement amounts, so that the readout base located between the second fastening mechanisms can be adjusted in horizontal position to achieve the effect of aligning with the central axis of the hemispherical shell.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] 1. The present invention collects data of assembly gap and electrode alignment in real time through the assembly process of the robotic arm assembly, and normalizes the collected data, uses the trained neural network model, predicts the model and outputs the prediction results, and then adjusts the assembly through the robotic arm assembly;

[0036] 2. The present invention converts the predicted adjustment parameter coordinates into the control signal of the robot arm assembly by adjusting the parameter coordinates, automatically adjusts the components in the x, y, and z directions, and completes the assembly accuracy adjustment of the hemispherical resonator and the readout base according to the prediction results and parameter coordinates output by the model, thereby meeting the calibration requirements;

[0037] 3. The present invention realizes data analysis and rapid calibration feedback functions of the hemispherical resonator and the readout base through the cooperation of model prediction and mechanical arm components. Through the coordinated work of various modules, automated assembly quality inspection and calibration are realized to meet the calibration values, reducing the operation steps and processes, thereby improving the assembly accuracy and production efficiency of the gyroscope, and further improving the quality factor and detection accuracy of the HRG. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a schematic block diagram of the process of the present invention;

[0039] Figure 2It is a schematic diagram of the structure of the present invention;

[0040] Figure 3 is a plan cross-sectional view of the clamping assembly;

[0041] Figure 4 for Figure 3 The enlarged schematic diagram of point A in the middle;

[0042] Figure 5 is a schematic diagram of a clamping state of a clamping assembly;

[0043] Figure identification: 1-mechanical arm assembly, 2-first fixed platform, 3-second fixed platform, 4-fixed cavity, 5-first positioning plate, 6-second positioning plate, 7-first elastic mechanism, 8-second fastening mechanism, 9-control motor, 10-driving rod, 11-horizontal push rod, 12-vertical cavity, 13-limiting ring, 14-second elastic mechanism, 15-pressure plate, 16-adjusting rod. DETAILED DESCRIPTION

[0044] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0045] Embodiment 1, as Figure 1 As shown, the assembly quality detection and calibration method based on the gyroscope assembly process disclosed by the present invention includes the following steps:

[0046] S1: Data collection and input, collecting assembly gap and electrode alignment data from the assembly line, including historical assembly data and real-time monitoring data, and normalizing them;

[0047] S2: Build LSTM neural network model;

[0048] S3: Model prediction, inputting the real-time monitored assembly gap and electrode alignment data into the model to obtain the predicted adjustment parameter coordinates;

[0049] S4: Model output, outputting the model assembly quality prediction result and adjustment parameter coordinates, converting the predicted adjustment parameter coordinates into control signals of the robot arm assembly, wherein the adjustment parameter coordinates include components in three directions: x, y, and z;

[0050] S5: The robotic arm assembly adjusts the first fastening mechanism for clamping the hemispherical shell in the vertical vector z direction and adjusts the second fastening mechanism for clamping the readout base in the horizontal vector x, y direction according to the control signal;

[0051] Wherein, the step S2 specifically includes the following sub-steps:

[0052] S201: extracting normalized assembly gap features and electrode alignment features;

[0053] S202: Use the normalized assembly gap features and electrode alignment features to train the LSTM model, and the output of the model is the adjustment parameter coordinates.

[0054] In step S1, the assembly gap is , the electrode alignment is , the normalization formula is:

[0055] ; .

[0056] Specifically, and are the maximum and minimum values ​​of the assembly clearance, respectively. It refers to the physical gap between the two components (hemispherical shell and readout base) during the assembly process of the gyroscope. The value range of the assembly gap is and between, It is the normalized assembly gap value calculated by linearly mapping the actual assembly gap of the sample with the minimum and maximum values ​​in the entire sample; and are the maximum and minimum values ​​of electrode alignment, respectively. It refers to the degree of alignment between the electrodes in the hemispherical shell and the electrodes in the readout base during assembly. The value range of electrode alignment is and between. It is the normalized electrode alignment value calculated by linearly mapping the actual electrode alignment of the sample with the minimum and maximum values ​​in the sample. and are the actual values ​​of the currently measured assembly gap and electrode alignment. and electrode alignment The measured values ​​may be at different magnitudes, and the electrode alignment may be a slight deviation in angle. Through normalization, these different magnitudes of data are converted to a unified scale so that the above values ​​can be compared and processed under the same standard.

[0057] In step S202, the hidden layer of the LSTM model is used to process the time series characteristics of the assembly gap and the electrode alignment, and the expression formula is:

[0058]

[0059] in, is the time step The hidden state of is the normalized input feature, and is the weight matrix, is the bias vector.

[0060] Specifically, is the time step The hidden state contains the information of the current time step and the accumulated information of all previous time steps. is the time step The hidden state of is updated by the information passed down from the previous time step through the forget gate and the input gate. It is a recursive weight matrix responsible for combining the hidden state of the previous time step with the input of the current time step to affect the hidden state of the current time step. is the input weight matrix used to combine the input features (normalized assembly gap and electrode alignment) at the current time step with the hidden state. It is the normalized assembly gap and electrode alignment data to ensure data consistency and comparability. is a bias vector, which is used to provide a basic activation level for the model and is regarded as the initial state of the model.

[0061] The hidden layer of the model captures the dynamic changes and long-term dependencies in the assembly process by processing the time series characteristics of the assembly gap and electrode alignment, enabling the model to more accurately predict the assembly quality and generate the corresponding adjustment parameter coordinates. The assembly gap and electrode alignment in the assembly process change over time. The model can capture the above changes and understand the temporal relationship between them. Finally, the adjustment parameter coordinates predicted by the model are directly used to control the robot arm components to achieve precise adjustment of the assembly process.

[0062] In step S4, the features of the hidden layer of LSTM are mapped to the output features, and the formula is: ,in, is the final hidden state of the hidden layer, are the predicted adjustment parameter coordinates.

[0063] Specifically, It is the predicted adjustment parameter coordinate, which is used to guide the robot arm component 1 to make adjustments. yes The state generated by the model after processing the input data of all time steps contains information about the entire sequence. is the output weight matrix, which is the weight matrix used to map the hidden state to the output feature space. is the output bias vector, which is used to provide the basic activation level for the output feature and determines the initial value of the output feature together with the output weight matrix. With the output weight matrix Multiply to achieve a linear transformation from the hidden state space to the output feature space. Added to the result of the linear transformation, it provides a basic activation level for the output feature, so that the value of the output feature can be flexibly adjusted. The final result is the predicted adjustment parameter coordinates, which are used to control the adjustment of the robotic arm component. This includes the components in the x, y, and z directions, which determine the specific position and posture of the robotic arm component in space.

[0064] In step S4, the difference between the predicted and actual values ​​of the assembly gap and electrode alignment of each sample is calculated, the squares of these differences are added, and then divided by the number of samples to obtain the final mean square error, thereby improving the prediction accuracy. The formula is:

[0065]

[0066] in, is the sample size, and They are The predicted values ​​of assembly gap and electrode alignment for samples, and It is The actual values ​​of assembly gap and electrode alignment of the samples, and are the minimum and maximum values ​​of the assembly clearance, and are the minimum and maximum values ​​of electrode alignment.

[0067] Specifically, the prediction errors of assembly gap and electrode alignment are first calculated, namely

[0068] and , and then divide the error by the difference between the maximum and minimum values ​​of each feature to normalize it. The normalized error squares are then added together to get the total error square sum. Finally, the total error square sum is divided by the number of samples. , and get the mean square error . The smaller it is, the smaller the difference between the model's predicted value and the actual value is, and the better the model performance is. , can guide the adjustment and optimization of model parameters. Through normalization, Ability to balance the error weights of different features to ensure that all features have equal importance in model evaluation.

[0069] In step S4, the model outputs the prediction result formula of assembly quality as follows: ,in It is an LSTM neural network model.

[0070] Specifically, the working principle of the neural network model includes feature input, feature learning, nonlinear transformation and output generation. The neural network receives the normalized assembly gap and electrode alignment as input features. The neural network learns the complex mapping relationship between the input features and the assembly quality adjustment parameters through multiple layers (such as hidden layers) inside it. The neural network introduces nonlinearity through activation functions such as ReLU, Sigmoid or Tanh, so that the model can capture nonlinear relationships. The output layer of the neural network generates the final prediction result , and obtain the adjustment parameters based on the input feature prediction. During the training process, the mean square error is used Or other loss functions to measure the difference between the predicted value and the actual value, and then use the back propagation algorithm to calculate the gradient of the loss function with respect to the network parameters to guide the update of the network parameters. Use gradient descent or other optimization algorithms to update the network parameters to minimize the loss function. Finally, through multiple iterative training, the neural network gradually learns the optimal mapping relationship between the input features and the output adjustment parameters.

[0071] In step S4, the coordinate formula of the adjustment parameter predicted by the output model is: ,in, , , are the components of the adjustment parameters in the three directions.

[0072] Specifically, and Indicates the adjustment amount in the horizontal direction. In the present application, the control motors 9 on the left and right sides are used as The direction of the control motor 9 is used as direction, the lifting and lowering of the clamping assembly According to the output adjustment parameters, the displacement direction and displacement amount of the control motor 9 are controlled, so as to adjust the horizontal position and longitudinal position of the sensitive base, ensure that the assembly of the axis of the hemispherical shell and the sensitive base meets the assembly requirements, and then perform welding connection.

[0073] An assembly quality inspection and calibration system based on the assembly quality inspection and calibration method in the gyroscope assembly process, such as Figure 2-5As shown, it includes a robotic arm assembly 1 and a clamping assembly, the clamping assembly is arranged at the control end of the robotic arm assembly 1, the clamping assembly includes a first fixed platform 2 and a second fixed platform 3, the first fixed platform 2 is a columnar structure with a bottom opening and a fixed cavity 4, the lower end of the fixed cavity 4 is provided with several first fastening mechanisms for fixing the hemispherical shell, the second fixed platform 3 is sleeved on the lower end of the first fixed platform 2, the bottom outer wall of the first fixed platform 2 is surrounded by a first positioning plate 5, the top inner wall of the second fixed platform 3 is surrounded by a second positioning plate 6, the first positioning plate 5 and the second positioning plate 6 are connected by a first elastic mechanism 7, the inner wall of the second fixed platform 3 is provided with multiple second fastening mechanisms 8 for adjusting the readout base, and the multiple second fastening mechanisms 8 are each equipped with a corresponding control motor 9, and the control motor 9 is connected to the control unit of the robotic arm assembly 1.

[0074] Specifically, the robot arm assembly 1 is used to realize the functions of handling and assembly, including rotating, lifting and moving the hemispherical shell. First, the control end of the robot arm assembly 1 rotates to the transportation line of the hemispherical shell, aligns the center line of the clamping assembly with the central axis of the hemispherical shell, and moves the clamping assembly downward. At this time, the first elastic mechanism 7 is stretched until the first elastic mechanism 7 reaches the maximum stretching state, the bottom of the first fixed platform 2 touches the placement plane of the hemispherical shell, and the hemispherical shell enters the fixed cavity 4. The hemispherical shell is limitedly clamped by the first fastening mechanism, and the clamping assembly is moved up after the fixation is completed. The first elastic mechanism 7 restores the elastic deformation, and the first fixed platform 2 moves up to form a accommodating cavity with the second fixed platform 3 that can accommodate the readout base. Then the control end of the mechanical part assembly rotates to move the clamping assembly fixed with the hemispherical shell to the assembly area, where the readout base is placed. The clamping assembly moves downward, aligns and assembles the hemispherical shell and the readout base, and controls each control motor 9 through the calculated adjustment parameter coordinates in the three directions of x, y, and z, so that the readout base is adjusted in the horizontal direction, and the z-axis direction is controlled by moving the clamping assembly up and down, and finally the function of precise adjustment is achieved. It should be noted that during the up and down movement of the clamping assembly, the z-axis direction of the hemispherical shell and the readout base can be adjusted first by increasing the clamping force of the first fastening mechanism and reducing the clamping force of the second fastening mechanism 8, that is, the sliding fixation of the readout base and the second fastening mechanism 8, and then the full fastening is achieved after the assembly standard of the hemispherical shell and the readout base is met.

[0075] The first fastening mechanism includes a driving rod 10 and a horizontal push rod 11. The side wall of the first fixed platform 2 is provided with several vertical cavities 12 for the driving rod 10 to move. The bottom surface of the driving rod 10 is inclined. One end of the horizontal push rod 11 is provided with an inclined surface that matches the bottom surface of the driving rod 10. The other end of the horizontal push rod 11 passes through the vertical cavity 12 and the fixed cavity 4. A limiting ring 13 and a second elastic mechanism 14 are provided on the horizontal push rod 11. When the driving rod 10 moves downward, the horizontal push rod 11 will be pushed toward the fixed cavity 4. A pressing plate 15 is provided above the first fixed platform 2. The tops of several driving rods 10 pass through the top surface of the first fixed platform 2 and are connected to the pressing plate 15. The top of the pressing plate 15 is connected with an output mechanism through an adjustment rod 16 that passes through the robotic arm assembly 1.

[0076] Specifically, the working principle of the first fastening mechanism is that a longitudinally movable output mechanism is arranged in the control end of the mechanical arm assembly 1, and the lifting and lowering of the adjusting rod 16 is controlled by the output mechanism, and the lifting and lowering of the pressure plate 15 is driven by the adjusting rod 16. The pressure plate 15 is connected to the driving rod 10, thereby controlling the lifting and lowering of several driving rods 10. A limiting portion cooperating with the limiting ring 13 is arranged on the movable cavity of the horizontal push rod 11, and the second elastic mechanism is located between the limiting ring 13 and the limiting portion. When the driving rod 10 is descending, the inclined surface at the bottom of the driving rod 10 pushes the horizontal push rod 11 to move into the fixed cavity 4, thereby achieving the function of clamping and fixing the hemispherical shell. After completing the assembly and welding of the hemispherical shell and the reading base, the control end of the mechanical arm assembly 1 rotates to the finished product placement area, and by moving the driving rod 10 upward, the second elastic mechanism 14 restores the elastic deformation, and the limiting ring 13 is away from the limiting portion, so that the horizontal push rod 11 rebounds into the vertical cavity 12, loosens the fixation of the hemispherical shell, and then places the finished product in the placement area.

[0077] A plurality of the second fastening mechanisms 8 are arranged at equal intervals, and at least four of the second fastening mechanisms 8 are provided. The second fastening mechanisms 8 have the same structure as the first fastening mechanisms. Specifically, the second fastening mechanisms 8 are controlled by the corresponding control motors 9 to achieve independent adjustment. Each control motor 9 receives different control signals according to the unified regulation of the control signal, and achieves different movement amounts, so that the readout base located between the second fastening mechanisms 8 can be adjusted in horizontal position to achieve the effect of alignment with the central axis of the hemispherical shell. Four second fastening mechanisms 8 can achieve the adjustment of the readout base on the x-axis and y-axis, and more second fastening mechanisms 8 can achieve more precise adjustment settings to meet the requirements of precise adjustment.

[0078] Embodiment 2, based on embodiment 1, this embodiment proposes a specific working principle of the assembly quality detection and calibration method based on the gyroscope assembly process:

[0079] Assume 3 groups of samples, and obtain the corresponding data samples of assembly gap and electrode alignment according to the samples. The values ​​are as follows:

[0080] Sample No. Assembly gap (μm) Electrode alignment Actual adjustment parameter coordinates (x, y, z) 1 0.3 0.8 （0.1,0.2,0.05） 2 0.4 0.75 （0.1,0.2,0.05） 3 0.2 0.9 （0.1,0.2,0.05）

[0081] Assembly gaps can be observed and measured by optical instruments such as laser scanners, and electrode alignment can use high-precision cameras and image processing algorithms to detect the position of the electrodes and calculate the alignment, or use laser beams to determine the center line of the electrodes and measure their alignment. Electrode alignment is usually expressed in dimensionless form, ranging from 0 to 1, indicating the accuracy of the alignment. The closer the value is to 1, the higher the alignment.

[0082] First, determine the maximum and minimum values ​​of the assembly gap and electrode alignment. Through 100 sets of sample values, the maximum values ​​are obtained as follows:

[0083] =0.1

[0084] =0.5

[0085] =0.7

[0086] =1.0

[0087] Then normalize the data of sample 1:

[0088]

[0089]

[0090] The normalized data is used to train a neural network model to learn to predict adjustment parameter coordinates based on the normalized assembly gap and electrode alignment.

[0091] For example, a new sample with an assembly gap of 0.45 and an electrode alignment of 0.85 is normalized as follows:

[0092]

[0093]

[0094] The normalized values ​​are then fed into the trained neural network model, which outputs the predicted adjustment parameter coordinates. .

[0095] The predicted adjustment parameter coordinates are converted into control signals for the robot arm, and the robot arm makes precise adjustments based on the above signals. After the assembly is completed, the assembly quality is evaluated and compared with the model prediction results. If there is a deviation between the predicted results and the actual assembly quality, feedback data is collected and the model is optimized to improve the prediction accuracy.

[0096] Of course, the present invention may have many other embodiments. Without departing from the spirit and essence of the present invention, technicians familiar with the field may make various corresponding changes and modifications based on the present invention, but these corresponding changes and modifications should all fall within the scope of protection of the claims attached to the present invention.

Claims

1. Based on the assembly quality detection and calibration method during the gyroscope assembly process, it is characterized by: The following steps are involved: S1: Data collection and input, collecting assembly gap and electrode alignment data from the assembly line, including historical assembly data and real-time monitoring data, and normalizing them; S2: Build LSTM neural network model; S3: Model prediction, inputting the real-time monitored assembly gap and electrode alignment data into the model to obtain the predicted adjustment parameter coordinates; S4: Model output, outputting the model assembly quality prediction results and adjustment parameter coordinates, converting the predicted adjustment parameter coordinates into control signals of the robot arm assembly and the clamping assembly, wherein the adjustment parameter coordinates include components in the three directions of x, y, and z; S5: Move the clamping assembly fixed with the hemispherical shell to the assembly area, on which the readout base is placed, move the clamping assembly downward, align and assemble the hemispherical shell with the readout base, and control each control motor respectively through the components of the three directions of the adjustment parameter coordinates x, y, and z obtained by the calculation, so that the readout base is adjusted in the horizontal direction, and the z-axis direction is controlled by moving the clamping assembly up and down to achieve the function of precise adjustment; Wherein, the step S2 specifically includes the following sub-steps: S201: extracting normalized assembly gap features and electrode alignment features; S202: using the normalized assembly gap features and electrode alignment features to train the LSTM model, the output of the model being the adjustment parameter coordinates; In step S4, the difference between the predicted and actual values ​​of the assembly gap and electrode alignment of each sample is calculated, the squares of these differences are added, and then divided by the number of samples to obtain the final mean square error, thereby improving the prediction accuracy. The formula is: among, represents the mean square error, is the sample size, and They are The predicted values ​​of assembly gap and electrode alignment for samples, and It is The actual values ​​of assembly gap and electrode alignment of the samples, and are the minimum and maximum values ​​of the assembly clearance, and are the minimum and maximum values ​​of electrode alignment; In step S4, the model outputs the prediction result formula of assembly quality as follows: ,in represents the error value calculated from the normalized assembly gap and electrode alignment through the functional relationship, yes Neural network model, the output layer of the neural network generates the final prediction result , obtain the adjustment parameters based on the input feature prediction. During the training process, the mean square error is used to measure the difference between the predicted value and the actual value. Then the back propagation algorithm is used to calculate the gradient of the loss function with respect to the network parameters to guide the update of the network parameters. It is the normalized assembly gap value calculated by linearly mapping the actual assembly gap of the sample with the minimum and maximum values ​​in the entire sample. It is the normalized electrode alignment value calculated by linearly mapping the actual electrode alignment of the sample with the minimum and maximum values ​​in the sample; In step S4, the adjustment parameter coordinates predicted by the model are calculated and output. The specific formula is: ,in, Represents the change vector of the adjustment parameter coordinates, , , are the components of the adjustment parameters in three directions; In step S5, a mechanical arm assembly (1) is included, wherein a clamping assembly is provided at a free end of the mechanical arm assembly (1), wherein the clamping assembly comprises a first fixing platform (2) and a second fixing platform (3), wherein a plurality of second fastening mechanisms (8) for adjusting a readout base are provided on an inner wall of the second fixing platform (3), wherein each of the plurality of second fastening mechanisms (8) is provided with a corresponding control motor (9), and the control end of the mechanical arm assembly (1) is rotated to a transport line of the hemispherical shell so that a center line of the clamping assembly is aligned with a center axis of the hemispherical shell, and the clamping assembly is moved downward, at which time the first elastic mechanism (7) is stretched until the first elastic mechanism (7) reaches a maximum stretched state, the bottom of the first fixing platform (2) touches a placement plane of the hemispherical shell, and the hemispherical shell enters into the fixing cavity (4), and the hemispherical shell is moved downward. The first fastening mechanism clamps the hemispherical shell in a limited manner, and the clamping assembly moves downward to align and assemble the hemispherical shell with the readout base. The components of the three directions of the adjustment parameter coordinates x, y, and z are calculated to control each control motor (9) respectively, so that the readout base is adjusted in the horizontal direction, and the z-axis direction is controlled by moving the clamping assembly up and down, thereby finally achieving the function of precise adjustment. The second fastening mechanism (8) realizes the function of independent adjustment under the control of the corresponding control motor (9). Each control motor (9) receives different control signals according to the unified regulation of the control signal, and achieves different movement amounts, so that the readout base located between the second fastening mechanisms (8) can be adjusted in horizontal position to achieve the effect of aligning with the central axis of the hemispherical shell.

2. The assembly quality detection and calibration method based on the gyroscope assembly process according to claim 1 is characterized in that: In step S1, the assembly gap is , the electrode alignment is , the normalization formula is: ； in, is the maximum value of the assembly clearance, is the minimum value of the assembly clearance, It is the normalized assembly gap value calculated by linearly mapping the actual assembly gap of the sample with the minimum and maximum values ​​in the entire sample; is the maximum value of electrode alignment, is the minimum value of electrode alignment, It is the normalized electrode alignment value calculated by linearly mapping the actual electrode alignment of the sample with the minimum and maximum values ​​in the sample.

3. The assembly quality detection and calibration method based on the gyroscope assembly process according to claim 1 is characterized in that: In step S202, the hidden layer of the LSTM model is used to process the time series characteristics of the assembly gap and the electrode alignment, and the expression formula is: in, is the time step The hidden state of is the normalized input feature, and is the weight matrix, is the bias vector, is the normalized value of the assembly clearance, is the normalized value of electrode alignment, It is a neural network structure that processes time series. yes The hidden state at the previous moment is passed as input to the calculation at the current moment to connect the information of the previous and next time steps.

4. The assembly quality detection and calibration method based on the gyroscope assembly process according to claim 1 is characterized in that: In step S4, the features of the hidden layer of LSTM are mapped to the output features, and the formula is: ,in, is the final hidden state of the hidden layer, are the predicted adjustment parameter coordinates, is the output weight matrix, which is the weight matrix used to map the hidden state to the output feature space. It is the output bias vector, which is used to provide the basic activation level for the output features and determines the initial value of the output features together with the output weight matrix.

5. An assembly quality inspection and calibration system using the assembly quality inspection and calibration method based on the gyroscope assembly process according to any one of claims 1 to 4, characterized in that: The invention comprises a mechanical arm assembly (1) and a clamping assembly, wherein the clamping assembly is arranged at the control end of the mechanical arm assembly (1), and the clamping assembly comprises a first fixing platform (2) and a second fixing platform (3). The first fixing platform (2) is a columnar structure with an opening at the bottom and provided with a fixing cavity (4). The lower end of the fixing cavity (4) is provided with a plurality of first fastening mechanisms for fixing a hemispherical shell. The second fixing platform (3) is sleeved on the lower end of the first fixing platform (2). A first positioning plate (5) is arranged around the bottom outer wall of the first fixing platform (2). A second positioning plate (6) is arranged around the top inner wall of the second fixing platform (3). The first positioning plate (5) and the second positioning plate (6) are connected via a first elastic mechanism (7). A plurality of second fastening mechanisms (8) for adjusting a readout base are arranged on the inner wall of the second fixing platform (3). The plurality of second fastening mechanisms (8) are each equipped with a corresponding control motor (9). The control motor (9) is connected to a control unit of the mechanical arm assembly (1).

6. The assembly quality inspection and calibration system according to claim 5, characterized in that: The first fastening mechanism comprises a driving rod (10) and a horizontal push rod (11); the side wall of the first fixed platform (2) is provided with a plurality of vertical cavities (12) for the driving rod (10) to move; the bottom surface of the driving rod (10) is inclined; one end of the horizontal push rod (11) is provided with an inclined surface that matches the bottom surface of the driving rod (10); the other end of the horizontal push rod (11) passes through the vertical cavity (12) and the fixed cavity (4); a limit ring (13) and a second elastic mechanism (14) are provided on the horizontal push rod (11); when the driving rod (10) moves downward, the horizontal push rod (11) is pushed toward the fixed cavity (4); a pressure plate (15) is provided above the first fixed platform (2); the tops of the plurality of driving rods (10) pass through the top surface of the first fixed platform (2) and are connected to the pressure plate (15); the top of the pressure plate (15) is connected to an output mechanism via an adjustment rod (16) that passes through the mechanical arm assembly (1).

7. The assembly quality inspection and calibration system according to claim 5, characterized in that: A plurality of the second fastening mechanisms (8) are arranged at equal intervals, at least four of the second fastening mechanisms (8) are provided, and the second fastening mechanisms (8) have the same structure as the first fastening mechanisms.

Images