A spare parts press fitting system and method based on deep learning algorithm

Abstract

The present invention provides a spare parts press-fitting system based on a deep learning algorithm, including a mechanical part, a control part and a software part. The system can accurately and stably control the press-fitting process, and provide real-time feedback on the press-fitting results. Due to its characteristic of predictably adjusting the press-fitting pressure, the deformation of the press-fitted parts can be reduced as much as possible. The present invention also provides a monitoring method for the press-fitting process of spare parts based on a statistical algorithm, which can screen out parts that are too tight or too loose during the press-fitting process, thereby improving the press-fitting quality. The present invention also provides a method for determining the press-fitting results of spare parts based on mechanical characteristics, which can accurately determine the results and quality of the press-fitting. The present invention also provides a method for predictively adjusting the press-fitting pressure during the press-fitting process of spare parts based on a deep learning algorithm, which can optimize the press-fitting process and reduce overpressure and the deformation caused by it.

Description

Technical Field

[0001] The present invention relates to the technical field of spare parts press-fitting, and in particular to a spare parts press-fitting system based on a deep learning algorithm and a monitoring and feedback method for a press-fitting process. Background Art

[0002] In the field of press-fitting of spare parts, hydraulic press-fitting machines are traditionally used to press-fit spare parts. Traditional hydraulic press-fitting machines are machines that use liquid as a working medium to transfer energy to achieve press-fitting. However, the pressure of traditional hydraulic presses is unstable and the standards are vague. The quality of press-fitting cannot be accurately judged and can only be judged by external dimension measurement and workers' perception. Due to the difference in tolerances of spare parts, overpressure may occur, resulting in product deformation. Traditionally, to accurately determine whether the spare parts have been press-fitted, secondary measurements are required. Traditionally, complex methods are required to determine the crimping quality of press fittings, which further increases the cost of press fittings.

[0003] In addition, for different products, workers traditionally rely on manual debugging and trial and error to obtain production standard parameters, and the results obtained are not accurate. At the same time, there is no feedback during the entire press-fitting process, and the quality of the finished product needs to be judged by the naked eye. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the above-mentioned existing defects, provide a parts press-fitting system based on a deep learning algorithm, a monitoring method for the parts press-fitting process based on a statistical algorithm, a method for determining the press-fitting result based on mechanical characteristics, and a method for predictive pressure adjustment during the parts press-fitting process based on a deep learning algorithm; the press-fitting process can be monitored in real time, real-time pressure data can be calculated during the press-fitting process, and pressure can be adjusted predictively. And after the press-fitting is completed, the press-fitting result can be accurately and real-timely given.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions.

[0006] The present invention provides a spare parts press-fitting system based on a deep learning algorithm, which includes a mechanical part, a control part and a software part; the mechanical part includes a servo motor, a pressure sensor, and a press-fitting device; the control part is developed based on a programmable logic controller; the software part includes an all-in-one machine, an interactive interface, an external interface and an algorithm module.

[0007] The control part is connected to the mechanical part, can send control signals, and obtain distance data of the servo motor and pressure data of the pressure sensor.

[0008] The software part is developed based on the all-in-one machine and communicates with the control part through Ethernet. It can set the press-fitting pressure and press-fitting speed of the control part and obtain the pressure-distance data of the control part.

[0009] The software part provides a human-computer interaction interface, and can call historical data to train an algorithm model. The algorithm module can be called to calculate the real-time pressure-distance data, and the calculation results are fed back to the control module and the external interface and the data is saved. The external interface can provide the results to the external system.

[0010] The system can accurately and stably control the press-fitting process, provide real-time feedback on the press-fitting results, and minimize the deformation of the press-fitted parts due to its ability to predictably adjust the press-fitting pressure.

[0011] The monitoring method of the spare parts press-fitting process based on statistical algorithms provided by the present invention is characterized in that, based on the central limit law and regression algorithm in statistics, the standardized historical press-fitting data is calculated, and the process control boundary is calculated and established; and during the press-fitting process, the boundary is used to monitor the real-time pressure-distance data, which specifically includes a boundary calculation process and a real-time monitoring process.

[0012] The purpose of the boundary calculation process is to establish the process control boundary. The purpose of the real-time monitoring process is to monitor the press-fitting process in real time. If there is any abnormality in the process, timely reminders and feedback to the superior system are given to avoid unqualified products from continuing to flow to the next process, wasting manpower and material resources.

[0013] The specific process is as follows:

[0014] 1. Boundary calculation process:

[0015] (1) The user sets the press-fitting speed and pressure during the press-fitting process through the human-computer interaction interface;

[0016] (2) The user uses the part press-fitting of the boundary size to obtain the pressure-distance data of the corresponding part press-fitting process;

[0017] (3) The algorithm module performs standardization processing on the pressure-distance data of the boundary size parts;

[0018] (4) The algorithm module generates process control boundaries based on the standardized boundary size part data using the regression algorithm in statistics.

[0019] 2. Real-time monitoring process:

[0020] (1) The user sets the running speed and pressure value of the press-fitting process through the human-computer interaction interface;

[0021] (2) Users use parts of any size for press fitting;

[0022] (3) The algorithm module obtains real-time pressure-distance data and standardizes the data;

[0023] (4) The algorithm module compares the standardized real-time pressure-distance data with the process control boundary. If the pressure-distance data exceeds the process control boundary, the part press fit is judged to be unqualified.

[0024] This method can screen out parts that are too tight or too loose during the press-fitting process, thereby improving the press-fitting quality.

[0025] The present invention provides a method for determining the results of press-fitting of spare parts based on mechanical characteristics, characterized in that, based on the difference in the elastic coefficient k value (slope in pressure-distance data) during the press-fitting process and after the parts are press-fitted, the k value interval after the parts are press-fitted is obtained by calculating the standardized historical data; and after the press-fitting is completed, based on the k value interval, the relative stroke, press-fitting pressure, and overpressure of the actual press-fitting are calculated, specifically including the calculation process and use process of the k value interval. The result of the press-fitted parts is determined by the overall elastic coefficient K value and the stroke. The method can accurately determine the results and quality of the press-fitting.

[0026] The specific process is as follows:

[0027] 1. Calculation process of k value interval:

[0028] (1) The user uses a certain number of press-fitted parts to repeat the press-fitting process;

[0029] (2) The algorithm module obtains the pressure-distance data of the corresponding parts press fit and performs standardization processing;

[0030] (3) The algorithm module calculates the k value of the press-fitted part and establishes the k value range;

[0031] 2. The process of using the k value interval:

[0032] (1) Users use parts of any size for press fitting;

[0033] (2) The algorithm module obtains the pressure-distance data of the corresponding part press-fitting process and performs standardization processing;

[0034] (3) The algorithm module uses the k value reaching the lower limit of the k value interval as the judgment method for press fitting completion, and calculates the relative stroke, press fitting pressure and overpressure of the press fitting completion point;

[0035] (4) The press-fitting results can be calculated immediately after the press-fitting is completed.

[0036] The result of the press-fitting completion is determined by the k value and the relative stroke. If the lower limit of the k value is reached within a certain relative stroke, the press-fitting is considered to be completed.

[0037] In addition, the present invention also provides a method for predictive pressure adjustment during the press-fitting process of spare parts based on a deep learning algorithm, which is characterized in that, based on the characteristics of deep learning that can learn and map logical relationships, a multi-layer network model is trained by acquiring the pressure-distance data of a large number of parts and the relative stroke and press-fitting pressure data of the actual press-fitting obtained in the above method for determining the press-fitting results of spare parts, and a mapping relationship between the pressure-distance process data and the relative stroke and press-fitting pressure of the actual press-fitting is established; and the model is used during the press-fitting process to predict the appropriate pressure to complete the press-fitting. The method predicts the press-fitting force required to complete the press-fitting of the product through the data during the press-fitting process, and dynamically completes the press-fitting. The method can complete the press-fitting with as little deformation as possible. Including training process and use process:

[0038] 1. Training process:

[0039] (1) Users use parts of any size for press fitting;

[0040] (2) The algorithm module obtains the pressure-distance data of the corresponding part press-fitting process and performs standardization processing;

[0041] (3) The algorithm module calculates the press-fitting completion point using the method in the above-mentioned method for determining the press-fitting result of spare parts;

[0042] (4) The algorithm module uses the pressure-distance data of several previous points as input, and the relative stroke and pressure of the press-fitting completion point as gt (result), trains a feedforward neural network, and obtains a deep learning model for predicting the press-fitting pressure;

[0043] 2. Usage process:

[0044] (1) Users use parts of any size for press fitting;

[0045] (2) During the press-fitting process, the algorithm module obtains the pressure-distance data and inputs it into the trained model in the training process to obtain the predicted pressure;

[0046] (3) The algorithm module sends the predicted pressure plus a certain pressure margin to the controller;

[0047] (4) The control part completes the press fitting with the set value;

[0048] (5) The obtained overpressure of the part is approximately equal to the set margin.

[0049] This method can optimize the press-fitting process and reduce overpressure and resulting deformation.

[0050] In summary, the present invention has the following beneficial effects:

[0051] 1. The press-fitting process is controllable. Through the monitoring method of the press-fitting process of spare parts based on statistical algorithms provided by the present invention, it is possible to screen out press-fitting that is too loose or too tight during the press-fitting process, and control the press-fitting quality.

[0052] 2. Accurate and instant feedback of press-fitting results. The method for determining the press-fitting results of spare parts based on mechanical characteristics provided by the present invention can accurately verify whether the press-fitting is completed, and immediately feedback the pressure used for press-fitting and the overpressure after the press-fitting is completed.

[0053] 3. Minimize the deformation of parts caused by high pressure. Through the method of predictably adjusting pressure during the press-fitting process of spare parts based on the deep learning algorithm provided by the present invention, the press-fitting pressure can be dynamically adjusted while ensuring the press-fitting results, and the press-fitting can be completed with the lowest possible pressure, thereby reducing the deformation of parts. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a schematic diagram of the module structure of the spare parts press-fitting system based on the deep learning algorithm proposed in the present invention.

[0055] Figure 2 This is a functional schematic diagram of the spare parts press-fitting system module based on the deep learning algorithm provided by the present invention.

[0056] Figure 3 Schematic diagram of a monitoring method for the press-fitting process of a turbine diffuser and a compressor casing based on a statistical algorithm in an embodiment.

[0057] Figure 4 It is a schematic diagram of a method for determining the press-fitting result of a turbine diffuser and a compression casing based on mechanical characteristics in an embodiment.

[0058] Figure 5 Schematic diagram of a method for predictively adjusting pressure during press-fitting of a turbine diffuser and a compression casing based on a deep learning algorithm in an embodiment.

[0059] Figure 6 The process control boundary diagram established for the press-fitting process of the turbine diffuser and compressor casing.

[0060] Figure 7 This is the distance-pressure data diagram for the press-fitting process of the turbine diffuser and compressor casing. DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] The purpose of the present invention is to provide a spare parts press-fitting system based on a deep learning algorithm, a spare parts press-fitting process monitoring method based on a statistical algorithm, a spare parts press-fitting result determination method based on mechanical characteristics, and a spare parts press-fitting process predictive pressure adjustment method based on a deep learning algorithm, to monitor the press-fitting process in real time, and to accurately and stably control the press-fitting process by setting pressure and speed. The press-fitting process can also refer to real-time pressure data to predictably adjust the pressure. After the press-fitting is completed, the press-fitting monitoring results can be accurately and real-time given.

[0063] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0064] Figure 1 The module schematic diagram of the spare parts press-fitting system based on deep learning provided by the present invention includes a mechanical part, a control part and a software part. The mechanical part includes a servo motor, a pressure sensor, and a press-fitting device. The control part is developed based on a programmable logic controller. The software part includes an all-in-one machine, an interactive interface, an external interface and an algorithm module.

[0065] Figure 2 This is a module function diagram of the deep learning-based spare parts press-fitting system provided by the present invention. The system operator can first set parameters through the software part, and the control part presses the spare parts according to the set parameters. The inductive sensor transmits the sensed pressure to the programmable logic controller of the control part, and the control part can also control the movement of the mechanical press-fitting part. The control part then transmits data such as distance pressure to the software part, and the software part uses the algorithm module to realize monitoring, result evaluation and predictable pressure adjustment.

[0066] Figure 3 The process diagram of the monitoring method for the press-fitting process of spare parts based on statistical algorithms proposed in the present invention is shown in Figure 1. The monitoring method is specifically described by taking the press-fitting process of the diffuser and the compressor shell in the turbine as an example. During the press-fitting process, the diffuser and the compressor shell need to be tightly pressed together without leaving any gaps, otherwise the working performance of the turbine will be affected. It includes the boundary calculation process S1 and the real-time monitoring process S2.

[0067] The purpose of the S1 boundary calculation process is to establish the process control boundary, which is divided into the following steps:

[0068] 1. The user sets the press-fitting speed and pressure of the diffuser and the compression shell through the human-computer interaction interface;

[0069] 2. The user uses the diffuser and the compressor shell of the boundary size for press fitting and obtains the pressure-distance data of the corresponding diffuser and the compressor shell;

[0070] 3. The algorithm module normalizes the pressure-distance data of the diffuser and compressor shell of the boundary size;

[0071] 4. The algorithm module uses the regression algorithm in statistics to control the boundary of the production process based on the standardized boundary size part data.

[0072] The established process control boundaries, i.e., envelopes, are as follows: Figure 6 As shown, curve 1 corresponds to the upper process control boundary, when the diffuser and the compression shell are too loosely matched. When the pressure-distance data during the press-fitting process exceeds this process control boundary, it indicates that the diffuser and the compression shell do not match in size during the press-fitting process, and the two parts cannot be tightly combined. Curve 2 corresponds to the lower process control boundary, when the diffuser and the compression shell are too tight. When the pressure-distance data during the press-fitting process exceeds this process control boundary, it indicates that the diffuser and the compression shell do not match in size during the press-fitting process, which can easily cause deformation or even damage to the parts.

[0073] The purpose of the S2 real-time monitoring process is to monitor the press-fitting process in real time. If there is any abnormality in the process, timely reminders and feedback to the superior system are given to avoid unqualified products from continuing to flow to the next process, wasting manpower and material resources. It is divided into the following steps:

[0074] 1. The user sets the press-fitting speed and pressure of the diffuser and the compression shell through the human-computer interaction interface;

[0075] 2. Users use parts of any size for press fitting;

[0076] 3. The algorithm module obtains the real-time pressure-distance data during the press-fitting process of the diffuser and the compressor shell, and standardizes the data;

[0077] 4. The algorithm module compares the obtained pressure-distance data with the process control boundary. If the pressure-distance data exceeds the process control boundary, it is judged that the press fit of the spare parts is unqualified.

[0078] Figure 7 The pressure-distance data diagram of the diffuser and the compressor shell during the press-fitting process is shown in Figure 1. Figure 7 From the data of curve 3, it can be seen that the distance-pressure data curve of the diffuser and the compressor shell does not exceed the process control boundary, indicating that the press fit of the diffuser and the compressor shell is qualified. The inflection point in the pressure-distance curve of the diffuser and the compressor shell is the completion point, that is, the overall elastic coefficient point of the diffuser and the compressor shell. The stroke corresponding to the part below the inflection point of curve 3 is the relative press fit stroke, and the pressure part corresponding to the part above the inflection point of curve 3 is the overpressure pressure.

[0079] Figure 4The present invention provides a schematic diagram of the process of determining the results of the press-fitting of spare parts based on mechanical properties, which specifically includes the calculation process and the use process of the k value interval. The press-fitting process of the diffuser and the compressor casing in the turbine is used as an example to specifically illustrate the determination method.

[0080] The calculation process of the k value range of the loudspeaker and the compression shell includes:

[0081] (1) The user uses a certain number of press-fitted diffusers and compression shells to repeat the press-fitting process;

[0082] (2) The algorithm module obtains the pressure-distance data of the corresponding diffuser and compressor shell press fit and performs standardization processing;

[0083] (3) The algorithm module calculates the k value of the press-fitted diffuser and compressor shell, and establishes the k value range.

[0084] The process of using the k-value range of the diffuser and compressor shell includes:

[0085] (1) The user can press-fit the diffuser and compressor shell of any size;

[0086] (2) The algorithm module obtains the pressure-distance data corresponding to the diffuser and the compressor shell during the press-fitting process and performs standardization processing;

[0087] (3) The algorithm module uses the k value reaching the lower limit of the k value interval as the judgment method for the completion of the press fitting, and calculates the relative stroke, press fitting pressure and overpressure of the press fitting completion point.

[0088] Figure 5 Schematic diagram of a method for predictive pressure adjustment during press-fitting of turbine diffuser and compressor casing based on a deep learning algorithm.

[0089] A method for predictively adjusting pressure during a press-fitting process is provided. The method for predictively adjusting pressure is specifically described by taking the press-fitting process of the diffuser and the compressor shell in a turbine as an example, including a training process W1 and a use process W2:

[0090] Training process W1:

[0091] (1) The user can press-fit the diffuser and the compressor shell of any size;

[0092] (2) The algorithm module obtains the pressure-distance data corresponding to the diffuser and the compressor shell during the press-fitting process and performs standardization processing;

[0093] (3) The algorithm module calculates the completion point of the diffuser and the compressor shell press fitting using the method in the above-mentioned spare parts result determination method;

[0094] (4) The algorithm module uses the pressure-distance data of several previous points as input, and the relative stroke and pressure of the press-fitting completion point as gt (result), trains a feedforward neural network, and obtains a deep learning model for predicting the press-fitting pressure.

[0095] Use process W2:

[0096] (1) The user can press-fit the diffuser and the compressor shell of any size;

[0097] (2) During the press-fitting process, the algorithm module obtains the pressure-distance data of the diffuser and the compressor shell, and inputs it into the trained model during the training process to obtain the predicted pressure;

[0098] (3) The algorithm module sends the predicted pressure plus a certain pressure margin to the controller;

[0099] (4) The control part completes the press-fitting of the diffuser and the compressor shell according to the set value.

[0100] Figure 7 The portion surrounded by the middle curve 4 is the half-stroke data and the press-fitting completion point data during the press-fitting process of the diffuser and the compressor shell, which is used to train the deep learning model, predict the pressure at the press-fitting point, dynamically configure the pressure, reduce the overpressure, and achieve predictable pressure adjustment.

Claims

1. A spare parts press-fitting system based on deep learning algorithm, including mechanical part, control part and software part; the mechanical part includes servo motor, pressure sensor and press-fitting device; the control part is developed based on programmable logic controller; the software part includes an integrated machine, interactive interface, external interface and algorithm module; it is characterized in that The control part is connected to the mechanical part, and can send control signals and obtain the distance data of the servo motor and the pressure data of the pressure sensor; the software part is developed based on the all-in-one machine, and communicates with the control part through Ethernet, and can set the press-fitting pressure and press-fitting speed of the control part, and obtain the pressure-distance data of the control part. Based on the central limit law and regression algorithm in statistics, the standardized historical press-fitting data is calculated and the process control boundary is calculated and established; and during the press-fitting process, the boundary is used to monitor the real-time pressure-distance data, which specifically includes the boundary calculation process and the real-time monitoring process; The boundary calculation process includes: (1) The user sets the press-fitting speed and pressure during the press-fitting process through the human-computer interaction interface; (2) The user uses the part press-fitting of the boundary size to obtain the pressure-distance data of the corresponding part press-fitting process; (3) The algorithm module performs standardization processing on the pressure-distance data of the boundary size parts; (4) The algorithm module generates process control boundaries based on the standardized boundary size part data using a regression algorithm in statistics; The real-time monitoring process includes: (1) The user sets the running speed and pressure value of the press-fitting process through the human-computer interaction interface; (2) Users use parts of any size for press fitting; (3) The algorithm module obtains real-time pressure-distance data and standardizes the data; (4) The algorithm module compares the standardized real-time pressure-distance data with the process control boundary. If the pressure-distance data exceeds the process control boundary, the part press fit is judged to be unqualified; Based on the difference in the elastic coefficient k value (the slope in the pressure-distance data) during and after the press-fitting of the parts, the k value interval after the press-fitting of the parts is obtained by calculating the standardized historical data; and after the press-fitting is completed, the relative stroke, press-fitting pressure, and overpressure of the actual press-fitting are calculated based on the k value interval, which specifically includes the calculation process and use process of the k value interval; The calculation process of the k value interval includes: (1) The user uses a certain number of press-fitted parts to repeat the press-fitting process; (2) The algorithm module obtains the pressure-distance data of the corresponding parts press fit and performs standardization processing; (3) The algorithm module calculates the k value of the press-fitted part and establishes the k value range; The process of using the k value interval includes: (1) Users use parts of any size for press fitting; (2) The algorithm module obtains the pressure-distance data of the corresponding part press-fitting process and performs standardization processing; (3) The algorithm module uses the k value reaching the lower limit of the k value interval as the judgment method for press fitting completion, and calculates the relative stroke, press fitting pressure and overpressure of the press fitting completion point; The result of the press-fitting completion is determined by the k value and the relative stroke. If the lower limit of the k value is reached within a certain relative stroke, the press-fitting is considered to be completed. Based on the characteristics of deep learning that can learn and map logical relationships, by acquiring the pressure-distance data of a large number of parts and the actual relative stroke and press-fitting pressure data of the press-fitting results obtained based on the spare parts press-fitting result judgment method, a multi-layer network model is trained to establish a mapping relationship between the pressure-distance process data and the actual relative stroke and press-fitting pressure of the press-fitting; and the pressure predicted by the model is used to complete the press-fitting during the press-fitting process.

2. The spare parts press-fitting system according to claim 1, characterized in that: The software part provides a human-computer interaction interface, through which the user can set the press-fitting pressure and speed of the control part, call historical data to train the algorithm model, call the algorithm module to calculate the real-time pressure-distance data, and feed back the calculation results to the control module and the external interface and save the data. The external interface can provide the results to the external system.

3. The spare parts press-fitting system based on deep learning algorithm according to claim 1 is characterized in that: The training process includes: (1) Users use parts of any size for press fitting; (2) The algorithm module obtains the pressure-distance data of the corresponding part press-fitting process and performs standardization processing; (3) The algorithm module calculates the press-fit completion point using the method in the spare parts press-fit result determination method; (4) The algorithm module uses the pressure-distance data of several previous points as input and the relative stroke and pressure of the press-fitting completion point as results to train a feedforward neural network and obtain a deep learning model for predicting the press-fitting pressure.

4. The spare parts press-fitting system based on deep learning algorithm according to claim 1 is characterized in that: The use process includes: (1) Users use parts of any size for press fitting; (2) During the press-fitting process, the algorithm module obtains the pressure-distance data and inputs it into the trained model in the training process to obtain the predicted pressure; (3) The algorithm module sends the predicted pressure plus a certain pressure margin to the controller; (4) The control part completes the press fitting with the set value.

Images