A control system for detecting automatic drinking fountains in laboratory animals

Abstract

This application relates to the field of automated detection technology and discloses a control system for detecting automatic drinking faucets in laboratory animals. The system includes a central control module and its subordinate signal acquisition, feature extraction, motion control, neural network prediction, and judgment execution modules. During the valve core opening phase of the needle-pulling mechanism, the system calculates the normalized flow resistance characteristic coefficient. During the needle-pulling reset phase, it adaptively adjusts the reset speed based on the flow resistance characteristics and extracts the mechanical hysteresis time difference feature. The neural network prediction module constructs a multi-dimensional input tensor based on the flow resistance characteristics, mechanical hysteresis characteristics, and pressure decay data, and outputs a dynamic qualified pressure threshold range for the current faucet. The judgment execution module employs dual-fuse logic, combining the comparison results of the measured pressure and the dynamic threshold with the physical boundary verification results of the mechanical hysteresis characteristics to generate the final sorting instruction. This invention enhances AI prediction and dual verification through physical mechanism features, improving the accuracy and reliability of detection.

Description

Technical Field

[0001] The present invention relates to the technical field of automated detection, and particularly to a control system for automatically detecting the drinking nozzles of laboratory animals. Background Art

[0002] As a key component of the laboratory water supply system, the sealing performance and opening sensitivity of the drinking nozzles of laboratory animals are directly related to experimental safety. Existing automated detection of nozzles usually adopts the air pressure decay method, that is, a mechanical device is used to push the valve core to open and inflate, and then the pressure change rate during the pressure holding or pressure relief process is monitored and compared with a preset fixed threshold to determine whether the product is qualified.

[0003] However, this single pneumatic detection method based on a fixed threshold has limitations in practical applications. The air source pressure in industrial sites usually fluctuates, and the machining and assembly tolerances of nozzle parts will cause individual differences in the physical flow area of the valve core. When using a fixed threshold for determination, it is impossible to distinguish whether the pressure change is caused by air source fluctuations, individual flow resistance differences or real leakage. Nozzles with a physically small opening but normal function are often misjudged as blocked due to slow pressure relief, while nozzles with a large opening may cover up minor leakage defects, resulting in limited detection accuracy.

[0004] In addition, the existing technology mainly focuses on the pressure values under static or steady state, lacking dynamic monitoring of the mechanical motion response characteristics of the valve core. The valve core of the nozzle may experience mechanical hysteresis or instantaneous jamming during the reset process due to spring fatigue or foreign object adhesion. If the valve core can return to its position and form a seal after a delay, conventional pressure detection methods often judge it as a qualified product. This detection method that only masks mechanical motion defects through airtightness data cannot effectively eliminate those hidden fault products that may cause difficult opening or continuous water flow in actual use.

[0005] Therefore, the present invention proposes a control system for automatically detecting the drinking nozzles of laboratory animals to solve the deficiencies of the existing technology. Summary of the Invention

[0006] Aiming at the deficiencies of the existing technology, the present invention provides a control system for automatically detecting the drinking nozzles of laboratory animals, which solves the problems of high detection misjudgment rate and missed detection of hidden mechanical faults in the existing nozzle detection technology due to the inability to eliminate the interference of air source fluctuations and individual differences, and the lack of monitoring of the mechanical motion characteristics of the valve core.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a control system for detecting automatic drinking nozzles in laboratory animals, applied to an automated detection device. The automated detection device includes a frame, a servo conveyor belt mounted on the frame, a detection station located on the conveying path of the servo conveyor belt, and a rejection cylinder located at the end of the servo conveyor belt. The servo conveyor belt conveys a tooling fixture carrying the drinking nozzle. The detection station is equipped with a fixing fixture, a sealing gas injection mechanism, and a needle-pulling mechanism. The fixing fixture clamps the drinking nozzle, the sealing gas injection mechanism injects test gas into the drinking nozzle, and the needle-pulling mechanism actuates the valve core of the drinking nozzle.

[0008] The control system includes a central control module, which includes:

[0009] The signal acquisition module is configured to acquire the absolute pressure signal of the sealing and gas injection mechanism and receive the motion status signal of the needle-pulling mechanism;

[0010] The feature extraction module is configured to calculate the normalized flow resistance characteristic coefficient based on the absolute pressure signal and extract the mechanical hysteresis time difference feature during the reset process of the needle-pulling mechanism.

[0011] The motion control module is configured to generate motion commands for the needle-picking mechanism and adaptively adjust the reset speed of the needle-picking mechanism according to the normalized flow resistance characteristic coefficient.

[0012] The neural network prediction module is configured to use a deep neural network model to output a dynamic qualified pressure threshold range based on the normalized flow resistance characteristic coefficient, mechanical hysteresis time difference characteristics, and pressure decay data.

[0013] The judgment execution module is configured to control the rejection cylinder sorting based on the comparison results of the measured pressure value and the dynamic qualified pressure threshold range, as well as the verification results of the mechanical hysteresis time difference characteristics and the physical safety range.

[0014] Preferably, the upper side of the frame is also provided with a stepper motor 1 and a stepper motor 2 located above the stepper motor 1, and the stepper motor 2 is connected to a robot arm; a fixing fixture 2 is provided on one side of the stepper motor 1, and an air blowing nozzle is provided on the side of the fixing fixture 2; the central control module is configured to control the stepper motor 1 and the stepper motor 2 to drive the robot arm to transfer the water nozzle to the fixing fixture 2 for pre-positioning, and to control the air blowing nozzle to clean the water nozzle with airflow, and then control the robot arm to transfer the cleaned water nozzle to the tooling fixture on the servo conveyor belt.

[0015] Preferably, the automated detection device further includes a waste bin located next to the rejection cylinder and a finished product guide chute located at the end of the servo conveyor belt; the central control module is configured to: when the current water nozzle is determined to be a defective product, control the rejection cylinder to push the water nozzle into the waste bin; when the current water nozzle is determined to be a qualified product, control the servo conveyor belt to transport the water nozzle to the finished product guide chute.

[0016] Preferably, the sealing and air-injection mechanism is configured to press down vertically to form a sealed connection with the air inlet end of the water nozzle; the needle-pulling mechanism is configured to move horizontally, and the needle-pulling mechanism includes a precision linear actuation unit equipped with a position feedback sensor; the signal acquisition module is configured to perform hardware clock synchronization calibration on the absolute pressure signal and the motion state signal of the needle-pulling mechanism to ensure that the timestamp of each pressure data point is strictly aligned with the timestamp of the position feedback data point.

[0017] Preferably, the feature extraction module is configured to calculate the first derivative of the pressure data with respect to time to obtain the instantaneous pressure drop rate during the dynamic opening phase when the needle-pulling mechanism pushes the valve core of the water nozzle to open, and define the ratio of the instantaneous pressure drop rate to the real-time absolute pressure value as the normalized flow resistance characteristic coefficient; the feature extraction module is further configured to statistically calculate the arithmetic mean, standard deviation and maximum value of the normalized flow resistance characteristic coefficient within the effective time window to generate an opening fingerprint feature vector.

[0018] Preferably, the motion control module is configured to read the arithmetic mean of the opening fingerprint feature vector before the needle-shifting mechanism performs the reset action; when the arithmetic mean is less than a preset flow resistance determination threshold, the motion control module determines that the current water nozzle has a large flow capacity and calls the speed parameter of the low-speed mode as the reset speed of the needle-shifting mechanism; when the arithmetic mean is greater than or equal to the flow resistance determination threshold, the speed parameter of the standard mode is called.

[0019] Preferably, the feature extraction module is configured to monitor the first derivative of the pressure data during the return verification phase of the reset action of the needle-pulling mechanism; when the pressure drop rate is detected to be zero or lower than a preset background noise threshold, the moment when the pressure drop rate is zero or lower than the preset background noise threshold is marked as the stop-fall inflection point moment; the feature extraction module calculates the time difference between the stop-fall inflection point moment and the start moment when the motion control module issues the reset command, and generates the mechanical hysteresis time difference feature.

[0020] Preferably, the feature extraction module further combines the mechanical hysteresis time difference feature with the current reset speed parameter to construct a mechanical hysteresis feature vector; the neural network prediction module is configured to obtain the pressure decay sequence of the steady-state leakage stage, the opening fingerprint feature vector generated in the dynamic opening stage, and the mechanical hysteresis feature vector generated in the back-trip verification stage; the neural network prediction module performs standardization processing and feature concatenation on the above data to construct a multidimensional input tensor.

[0021] Preferably, the deep neural network model loaded in the neural network prediction module includes a temporal feature extraction branch and a mechanism feature mapping branch; the output layer of the deep neural network model is configured as a regression output layer, used to output the dynamic qualified pressure threshold interval containing the minimum qualified pressure value and the maximum qualified pressure value, and the dynamic qualified pressure threshold interval dynamically floats with the change of the input multidimensional input tensor.

[0022] Preferably, the determination execution module is configured to execute dual circuit breaker determination logic: the first determination is an artificial intelligence compliance determination, whereby an artificial intelligence compliance signal is generated when the measured pressure value at the end of the steady-state leakage stage falls within the dynamic qualified pressure threshold range; the second determination is a mechanical fault determination, whereby a mechanical fault signal is generated when the mechanical hysteresis time difference characteristic exceeds a preset physical safety range, wherein the physical safety range is defined by a high-level time threshold for anti-jamming and a low-level time threshold for anti-loose connection; the determination execution module is configured to perform logical operations on the artificial intelligence compliance signal and the mechanical fault signal; the current faucet is determined to be a qualified product if and only if the artificial intelligence compliance signal is generated and no mechanical fault signal is generated; if the mechanical fault signal is generated, the current faucet is forcibly determined to be a defective product and a rejection instruction is generated, regardless of whether the artificial intelligence compliance signal exists.

[0023] This invention provides a control system for automatically detecting water taps in laboratory animals. It has the following beneficial effects:

[0024] 1. This invention calculates the normalized flow resistance characteristic coefficient through the feature extraction module, effectively eliminating the interference of initial pressure fluctuations of the air source in industrial sites on valve core opening identification; the motion control module uses this flow resistance characteristic for feedforward control, which can adaptively adjust the reset speed of the needle-pulling mechanism; for water nozzles with abnormal flow capacity, the system automatically reduces the reset speed and enters the sensitive mode, thereby significantly improving the detection resolution of hidden faults such as minor mechanical jamming of the valve core or spring fatigue during the return verification stage, solving the problem of missed detection of mechanical faults under the traditional fixed speed detection mode.

[0025] 2. This invention employs a neural network prediction module to construct a multi-dimensional input tensor containing flow resistance characteristics, mechanical hysteresis characteristics, and pressure decay data, thereby achieving deep learning prediction based on physical mechanism enhancement. This method can output a dynamic qualified pressure threshold range that matches the physical flow resistance characteristics and mechanical friction characteristics of each individual faucet. This effectively solves the technical problem in the traditional single fixed threshold judgment method, where the difference in the opening tolerance of individual faucets leads to the misjudgment of normal small opening slow decay as blockage, or the misjudgment of abnormal large opening fast decay as leakage.

[0026] 3. This invention establishes a dual-fuse determination logic in the determination execution module, which combines AI probabilistic prediction and mechanical deterministic verification, and gives the mechanical hysteresis time difference feature a mandatory veto power over the final sorting result; even if the measured pressure data meets the neural network prediction standard, once the mechanical response time of the valve core is detected to exceed the physical safety range, the system will directly trigger the fuse and determine that it is unqualified; this deterministic physical constraint mechanism effectively avoids the risk of misjudgment that may occur under extreme boundary conditions by simply relying on artificial intelligence algorithms, and ensures the reliability of the detection results. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the system architecture of the present invention;

[0028] Figure 2 This is a schematic diagram of the method flow of the present invention;

[0029] Figure 3 This is a schematic diagram illustrating the principle of detection timing and physical mechanism feature extraction in this invention;

[0030] Figure 4 This is a schematic diagram of the automatic drinking spout detection device for laboratory animals according to the present invention. Figure 1 ;

[0031] Figure 5 For the present invention Figure 4 Enlarged view of point A in the middle;

[0032] Figure 6 This is a schematic diagram of the automatic drinking spout detection device for laboratory animals according to the present invention. Figure 2 .

[0033] The components include: 1. Frame; 2. Water nozzle; 3. Stepper motor one; 4. Servo conveyor belt; 5. Stepper motor two; 6. Robot arm; 7. Tooling fixture; 8. Fixture one; 9. Sealing and air injection mechanism; 10. Needle-pulling mechanism; 11. Scrap bin; 12. Fixture two; 13. Air blowing nozzle; 14. Rejection cylinder; 15. Finished product guide chute.

[0034] 100. Central control module; 101. Signal acquisition module; 102. Feature extraction module; 103. Motion control module; 104. Neural network prediction module; 105. Decision execution module. Detailed Implementation

[0035] The technical solutions in 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.

[0036] See attached document Figure 1 - Appendix Figure 6 This invention provides a control system for automatically detecting water taps in laboratory animals. The system is applied to an automated detection device. The automated detection device includes a frame 1, with a stepper motor 3 mounted on the upper side of the frame 1. A second stepper motor 5 is mounted above the first stepper motor 3 via a connector. A robotic arm 6 is connected to the second stepper motor 5 and is used to perform grasping and transferring actions of the water tap 2. A servo conveyor belt 4 is also mounted on the frame 1. Multiple tooling fixtures 7 are spaced apart on the servo conveyor belt 4 to carry the water tap 2 to be tested.

[0037] A fixing fixture 2 12 is provided on one side of the stepper motor 3. The fixing fixture 2 12 is used to pre-position the water nozzle 2 before the robot arm 6 picks up the water nozzle 2. An air blowing nozzle 13 is provided on the side of the fixing fixture 2 12. The air blowing nozzle 13 is configured to clean the water nozzle 2 with airflow. A detection station is provided on the conveying path of the servo conveyor belt 4. A fixing fixture 8 is provided at the detection station. The fixing fixture 8 is configured to clamp the water nozzle 2 in the tooling fixture 7 during the detection process to prevent the water nozzle 2 from displacing when subjected to force.

[0038] At the testing station, a sealing gas injection mechanism 9 and a needle-pulling mechanism 10 are provided. The sealing gas injection mechanism 9 is located above the tooling fixture 7 and is configured to press down vertically to form a sealed connection with the air inlet end of the water nozzle 2 and inject test gas at a set pressure into the water nozzle 2. The needle-pulling mechanism 10 is located on the side of the tooling fixture 7 and is configured to move horizontally to push the valve core of the water nozzle 2 to actuate. The needle-pulling mechanism 10 includes a precision linear actuator unit, which is equipped with a position feedback or status monitoring sensor.

[0039] At the end of the servo conveyor belt 4, there is a rejection cylinder 14, a waste bin 11, and a finished product guide chute 15. The rejection cylinder 14 is configured to push the unqualified water nozzles 2 into the waste bin 11 according to the instructions of the control system, while the qualified water nozzles 2 enter the finished product guide chute 15 with the servo conveyor belt 4.

[0040] The control system includes a central control module 100, which is electrically connected to stepper motor 3, stepper motor 5, servo conveyor belt 4, sealing gas injection mechanism 9, needle picking mechanism 10, and rejection cylinder 14 to achieve coordinated control of the various actuators. The central control module 100 has a data processing unit internally or in communication with it, which is further divided into the following modules:

[0041] The signal acquisition module 101 is electrically connected to the pressure sensor installed in the pipeline of the sealing gas injection mechanism 9, and is configured to acquire the absolute pressure signal in the pipeline at a preset high frequency; the signal acquisition module 101 is also configured to receive the motion status signal of the needle-pulling mechanism 10, and to perform time synchronization calibration on the pressure signal and the motion status signal.

[0042] The feature extraction module 102 is configured to perform digital filtering on the raw pressure data acquired by the signal acquisition module 101. The feature extraction module 102 is also configured to calculate the first derivative of the pressure data with respect to time and calculate the normalized flow resistance characteristic coefficient based on the real-time pressure value. The feature extraction module 102 is also configured to monitor the response characteristics of the needle-pulling mechanism 10 during the reset process and calculate the mechanical hysteresis time difference characteristics.

[0043] The motion control module 103 is configured to generate advance and reset commands for the needle-picking mechanism 10. The motion control module 103 has an adaptive adjustment function, which can dynamically adjust the reset speed parameter or reset monitoring mode of the needle-picking mechanism 10 according to the flow resistance feature data output by the feature extraction module 102.

[0044] The neural network prediction module 104 is loaded with a pre-trained deep neural network model. The neural network prediction module 104 is configured to receive a multi-dimensional input tensor constructed from the flow resistance feature vector, mechanical hysteresis feature vector and pressure decay curve data output by the feature extraction module 102. Based on the multi-dimensional input tensor, the neural network prediction module 104 outputs the dynamic qualified pressure threshold range for the current tested faucet 2.

[0045] The judgment execution module 105 is configured to compare the measured pressure value with the dynamic qualified pressure threshold range output by the neural network prediction module 104; the judgment execution module 105 is also configured to independently verify whether the mechanical hysteresis time difference feature output by the feature extraction module 102 is within the preset physical safety range; the judgment execution module 105 generates a qualified or unqualified judgment signal based on the comprehensive comparison result, and controls the rejection cylinder 14 to perform the corresponding sorting action.

[0046] The detection process executed by the control system is divided into three consecutive stages in the time domain:

[0047] During the dynamic opening phase T1, the motion control module 103 controls the needle-pulling mechanism 10 to move forward to open the valve core of the water nozzle 2, and the feature extraction module 102 simultaneously extracts the flow resistance features that characterize the flow capacity of the physical opening of the valve core.

[0048] During the steady-state leakage stage T2, the needle-pulling mechanism 10 remains in the same position, and the signal acquisition module 101 continuously collects pressure decay data. This stage mainly reflects the sealing performance of the water tap 2.

[0049] During the return verification phase T3, the motion control module 103 controls the needle-pulling mechanism 10 to perform a reset action; this return verification phase T3 overlaps and runs parallel to the mechanical reset stroke of the needle-pulling mechanism 10 on the time axis; during this phase, the feature extraction module 102 captures the bottoming point of the pressure curve and calculates the mechanical hysteresis characteristics to verify the mechanical return performance of the valve core.

[0050] Through the coordinated operation of the above modules, the central control module 100 completes the identification of the physical opening degree of the water tap 2, the sealing test, and the verification of the mechanical motion performance within a single detection cycle.

[0051] This invention provides a control method for detecting automatic drinking fountains in laboratory animals, comprising the following steps:

[0052] S100, the control conveying unit conveys the tooling fixture 7 loaded with water nozzle 2 to the testing station, controls the sealing gas injection mechanism 9 to press down to establish a sealed connection with water nozzle 2, and fills the test gas to the preset initial pressure;

[0053] S200, enter the dynamic opening stage T1, control the needle-pulling mechanism 10 to push the valve core of the water nozzle 2 to open with a preset displacement, collect real-time pressure data and perform filtering preprocessing, calculate the normalized flow resistance characteristic coefficient, and generate the opening fingerprint feature vector.

[0054] S300, enter steady-state leakage stage T2, keep the position of the needle-pulling mechanism 10 unchanged, and continuously collect pressure decay curve data;

[0055] S400, enter the return verification stage T3, adaptively generate the reset control parameters of the pin-shifting mechanism 10 based on the opening fingerprint feature vector generated in S200, control the pin-shifting mechanism 10 to perform the reset action, and simultaneously monitor the bottoming point of the pressure curve to generate the mechanical hysteresis feature vector.

[0056] S500 constructs a multi-dimensional input tensor from the opening fingerprint feature vector, mechanical hysteresis feature vector, and pressure decay curve data, inputs it into a pre-trained neural network model, and outputs the dynamic qualified pressure threshold range for the current water tap 2.

[0057] S600 compares the measured pressure data with the dynamic qualified pressure threshold range and independently verifies whether the mechanical hysteresis characteristic vector is within the physical safety range. Based on the comprehensive judgment result, it controls the rejection cylinder 14 to perform the sorting action.

[0058] To further clarify the implementation of each technical aspect of the present invention, the following will provide a detailed description of the implementation of each functional module involved above and its internal processing flow.

[0059] See attached document Figure 1 In the specific implementation of the signal acquisition module 101, in order to ensure the capture of the transient fluid characteristics of the valve core of the water nozzle 2 under millisecond-level operation and to provide an accurate time reference for subsequent mechanical hysteresis verification, the signal acquisition module 101 adopts a high-frequency synchronous acquisition and digital filtering strategy. The operation flow of the signal acquisition module 101 specifically includes the following sub-steps:

[0060] S101, establish a high-frequency discrete pressure data sequence; the pressure sensor configured in the sealing gas injection mechanism 9 pipeline adopts a piezoelectric or strain gauge high-precision sensor element. This sensor is equipped with an analog-to-digital conversion unit, and the sampling accuracy of the analog-to-digital conversion unit is preferably set to 16 bits or 24 bits to distinguish minute pressure fluctuations; the system sets the sampling frequency. The frequency ranges from 1 kHz to 10 kHz; after the detection cycle is started, the signal acquisition module 101 operates according to the sampling frequency. The analog voltage signal from the pressure sensor is continuously read and converted into a digital signal to generate the original discrete pressure sequence. ,in, For indexes of discrete time series, The higher sampling frequency configuration is to meet the need to capture rapidly changing pressure gradient signals during the dynamic start-up phase T1 and the back-pass verification phase T3.

[0061] S102 performs hard clock synchronization calibration of multi-source signals; since the flow resistance characteristic analysis of the system depends on pressure changes, and the mechanical hysteresis characteristic analysis depends on the timing relationship between pressure changes and the physical position of the needle-shifting mechanism 10, it is necessary to eliminate the transmission delay and clock drift between pressure data and position data; the central control module 100 is configured with a distributed clock mechanism based on a bus communication protocol; during the acquisition process, the timing when the signal acquisition module 101 reads the pressure data and the timing when the motion control module 103 reads the position feedback data of the needle-shifting mechanism 10 are both triggered by the same global master clock signal.

[0062] In practice, each time the original discrete pressure sequence is collected... All were given a precise timestamp The timestamp Position status data of the needle-shifting mechanism 10 The timestamps are strictly aligned, and the time deviation between them is controlled within a single sampling period. Within this range; this mechanism based on hard clock synchronization forms the physical basis for the subsequent accurate calculation of mechanical hysteresis time difference, avoiding the uncertainty delay caused by software polling mechanism.

[0063] S103, digital filtering preprocessing is performed on the raw pressure data; because the gas medium generates turbulence noise and high-frequency electromagnetic interference during high-speed flow and valve opening and closing, these noises will seriously affect the stability of subsequent first-order derivative calculations of the pressure data; therefore, before performing feature calculations, the feature extraction module 102 preprocesses the raw discrete pressure sequence through the signal acquisition module 101 or the internal preprocessing unit. Perform moving average filtering.

[0064] The moving average filtering algorithm smooths random noise by setting a fixed-length sliding window and averaging the data within the window; the resulting smoothed pressure sequence... The calculation formula is as follows:

[0065] ;

[0066] In the formula: Indicates time Pressure value after filtering; Indicates time The raw pressure values ​​collected; This indicates the width of the sliding window. It is a positive integer.

[0067] Window width The value is selected based on the bandwidth ratio of the system noise frequency to the effective signal frequency; in this embodiment, The value range is set to 5-20; this range effectively filters out high-frequency fluid noise while avoiding excessive phase delay on the true inflection point characteristics of the pressure curve, thus ensuring the real-time performance of subsequent bottoming-out inflection point identification. The specific code or circuit implementation of the moving average filter is a standard technique in digital signal processing and will not be described further here. The smoothed pressure sequence obtained after processing by S103... It will be transmitted to the feature extraction module 102 as the basic data source for subsequent calculation of normalized flow resistance characteristic coefficients and identification of system state switching.

[0068] In the specific implementation of the feature extraction module 102, in order to solve the interference caused by the initial pressure fluctuation of the air source and the machining tolerance on the opening degree recognition, the feature extraction module 102 constructs a physical model based on normalized flow resistance, and accurately quantifies the physical flow capacity of the valve core by calculating the dimensionless flow resistance characteristic coefficient. The operation process of the feature extraction module 102 in the dynamic opening stage T1 specifically includes the following sub-steps:

[0069] S201, calculate the instantaneous pressure drop rate; when the needle-pulling mechanism 10 contacts and pushes the valve core of the water nozzle 2 to open, the gas in the sealed pipeline is rapidly released outward through the valve core opening, causing the pipeline pressure to show a non-linear decreasing trend; the feature extraction module 102 receives the smoothed pressure sequence transmitted by the signal acquisition module 101. Then, a first-order discrete-time difference operation is performed on it to obtain the instantaneous voltage drop rate at each moment. Instantaneous voltage drop rate The calculation formula is as follows:

[0070] ;

[0071] In the formula: Indicates time The instantaneous pressure drop rate, usually expressed in kPa / s; Indicates time Smooth pressure value; Indicates time Smooth pressure value; This represents the sampling period, and its value is equal to the sampling frequency. The reciprocal of, that is .

[0072] S202, construct a normalized flow resistance characteristic model; in actual industrial settings, the initial pressure of the compressed air source... Fluctuations are often observed; according to fluid mechanics principles, for the same physical opening size, the higher the driving pressure, the faster the gas outflow velocity, exhibiting a pressure drop rate. The larger the voltage drop rate, the greater the voltage drop rate; if the voltage drop rate is used directly... Using this method to determine the opening size can confuse the variables of "pressure fluctuation" and "opening change," leading to identification errors.

[0073] Therefore, this invention introduces a normalized flow resistance characteristic coefficient. This coefficient is based on the quasi-steady-state assumption of gas flow, normalizing the instantaneous pressure drop rate with respect to the current absolute pressure; normalized flow resistance characteristic coefficient. The calculation formula is as follows:

[0074] ;

[0075] In the formula: Indicates time The normalized flow resistance characteristic coefficient, whose physical meaning approximately characterizes the aerodynamic admittance or equivalent flow area of ​​the valve core opening at the current moment. Represents the absolute value of the instantaneous voltage drop rate; Indicates time The absolute pressure value.

[0076] Based on the above calculation model, when the initial pressure of the gas source... When drift occurs, the numerator term With denominator They will change proportionally in the same direction, and the ratio of the two will be... Maintaining relative stability; this technical feature enables the system to isolate the gas source pressure variable and extract the essential features that are only related to the physical structure of the valve core (opening size, flow channel shape), thus achieving decoupling from environmental interference.

[0077] S203, generate the opening fingerprint feature vector; the feature extraction module 102 defines the effective time window of the dynamic opening phase T1, which starts when the voltage drop rate exceeds the trigger threshold and ends when the voltage drop rate change rate tends to stabilize; within this effective time window, the system processes a series of calculated features... Statistical analysis is performed on the values ​​to extract multi-dimensional statistics in order to construct an openness fingerprint feature vector. Openness fingerprint feature vector Its construction form is as follows:

[0078] ;

[0079] In the formula: Indicates all within the valid time window The arithmetic mean of the values ​​is used to reflect the average flow capacity of the valve core during the opening process; Indicates the effective time window The standard deviation of the value is used to reflect the smoothness of the opening process and the airflow disturbance. Indicates the effective time window The maximum value is used to reflect the maximum instantaneous opening degree reached by the valve core.

[0080] The feature extraction module 102 will generate the openness fingerprint feature vector. The vector is temporarily stored in the system memory. On the one hand, it is passed to the motion control module 103 as an input parameter to generate adaptive feedforward control commands in the subsequent S400 step. On the other hand, it is passed to the neural network prediction module 104 as a physical mechanism feature to participate in the prediction calculation of dynamic threshold. Through the processing of S201 to S203, the system completes the transformation from raw pressure data to physical mechanism features.

[0081] See attached document Figure 3 In the specific implementation of the return verification stage T3, this invention adopts a feedforward control mechanism that adaptively adjusts the subsequent control strategy based on the preceding detection results, and combines time-domain feature analysis technology to extract hysteresis features reflecting the mechanical motion performance of the valve core; this process is mainly completed collaboratively by the motion control module 103 and the feature extraction module 102, and specifically includes the following sub-steps:

[0082] S401, Generate adaptive reset control parameters; Motion control module 103 receives the opening fingerprint feature vector generated by feature extraction module 102 in the preceding dynamic opening phase T1. The motion control module 103 has a pre-installed parameter mapping unit, which stores a mapping table or functional relationship between flow resistance characteristics and reset speed.

[0083] Motion control module 103 analyzes the opening fingerprint feature vector arithmetic mean And based on this value, determine the current operating condition category of faucet 2; when Less than the preset flow resistance threshold When this occurs, it indicates that the water nozzle 2 is exhibiting a large flow capacity, which may indicate insufficient spring preload or a short stroke. In this case, the motion control module 103 retrieves the speed parameters for the low-speed, high-resolution mode. The target speed for this reset action; conversely, when Greater than or equal to the flow resistance determination threshold When this occurs, it indicates that the flow characteristics of water nozzle 2 are normal, and the motion control module 103 retrieves the speed parameters of the standard mode. Determined reset speed Generated through the following mapping logic:

[0084] ;

[0085] In the formula: This indicates the set speed of the current reset action of the needle-shifting mechanism 10; This represents a preset parameter mapping function; This represents the critical threshold constant that distinguishes different flow resistance characteristics. and All are preset speed constants, and .

[0086] Through this step, the system can dynamically match the optimal mechanical action parameters for each individual water tap based on its physical opening characteristics, thereby improving the detection resolution for abnormal individuals while ensuring detection efficiency.

[0087] S402, execute the reset action and monitor the pressure response; motion control module 103 sends a reset command to needle-picking mechanism 10, driving needle-picking mechanism 10 at a determined speed. The retraction action is performed, allowing the valve core of the water nozzle 2 to reset and close under its own spring force; while the needle-pulling mechanism 10 is in motion, the feature extraction module 102 continuously receives the real-time pressure data transmitted by the signal acquisition module 101 and monitors the changing trend of the pressure curve.

[0088] During this process, the system records the precise moment when the motion control module 103 issues the reset command, and records it as the reset start moment. This moment is determined based on the global synchronization clock.

[0089] S403 captures the inflection point of the pressure curve; during the transition of the valve core from the open to the closed state, the amount of gas leakage in the sealed pipeline gradually decreases, causing the pressure drop rate to gradually slow down; when the valve core is fully closed and the seal is good, the pipeline pressure will stop dropping and tend to stabilize (or only a very small amount of background leakage exists); the feature extraction module 102 extracts the first derivative of the pressure data (i.e., the pressure drop rate). Real-time monitoring can be performed.

[0090] Feature extraction module 102 identifies the moment when the pressure decay rate suddenly changes to zero or falls below a preset background noise threshold, and defines it as the moment of bottoming out. The following mathematical criterion is satisfied at this moment:

[0091] ;

[0092] In the formula: This indicates the turning point when the price stops falling; Indicates time The rate of change of pressure; This represents the minimum threshold value at which the pressure tends to reach a steady state.

[0093] S404, calculate the mechanical hysteresis feature vector; feature extraction module 102 is based on the reset start time. At the turning point of the decline Calculate the hysteresis time difference that reflects the mechanical response delay of the valve core. This hysteresis time difference is a macroscopic manifestation of the combined effects of valve core friction, spring restoring force, and fluid resistance; the calculation formula is as follows:

[0094] ;

[0095] Given that the hysteresis time difference is affected by the reset speed, in order to provide complete physical context information to the subsequent neural network model, the feature extraction module 102 will use the calculated hysteresis time difference... With the current reset speed Combine to construct mechanical hysteresis feature vectors Mechanical hysteresis eigenvector The expression is as follows:

[0096] ;

[0097] Feature extraction module 102 outputs the mechanical hysteresis feature vector. This vector is used on the one hand to determine the physical fault circuit verification of the execution module 105, and on the other hand, it is transmitted to the neural network prediction module 104 as part of the input tensor. Through the above steps, the system completes the quantitative detection of the mechanical motion performance of the valve core of the water tap 2 without adding extra detection stations and time cycles.

[0098] In the specific implementation of the neural network prediction module 104, to address the problem that single pressure time-series data cannot fully reflect the individual differences between the two faucets, leading to insufficient prediction accuracy, this invention employs multi-source feature fusion technology. This maps the physical mechanism features extracted in the preceding steps to a high-dimensional feature space, constructs a physical mechanism-enhanced input tensor, and outputs a dynamically adapted judgment threshold through a deep learning model. The specific operation flow of the neural network prediction module 104 includes the following sub-steps:

[0099] S501, construct a physical mechanism-enhanced input tensor; at the end of the steady-state leakage phase T2 of the detection cycle, the neural network prediction module 104 simultaneously acquires data sources in three dimensions:

[0100] The first data source is the pressure decay sequence collected during the steady-state leakage phase T2, denoted as... ;

[0101] The second data source is the opening fingerprint feature vector generated by the feature extraction module 102 during the dynamic opening phase T1. ;

[0102] The third data source is the mechanical hysteresis feature vector generated by the feature extraction module 102 during the back-trip verification stage T3. .

[0103] To eliminate the interference of different physical dimensions (pressure unit kPa, time unit ms, flow resistance coefficient dimensionless) on the neural network weight update, the neural network prediction module 104 first performs standardization processing on the above data; the processed data are denoted as follows: , and .

[0104] Subsequently, the neural network prediction module 104 performs a feature concatenation operation; because It is a time-series vector, and and These are low-dimensional statistical feature vectors. In this embodiment, a multi-channel input architecture or vector copying expansion strategy is used for fusion. In this embodiment, a physical mechanism-enhanced input tensor is constructed using feature concatenation. Its logical expression is as follows:

[0105] ;

[0106] In the formula: This represents the comprehensive input data fed into the neural network model; This indicates tensor splicing or feature channel overlay operation; This indicates the current sealing condition of faucet 2; It characterizes the physical opening capability of the water tap 2; The mechanical motion friction characteristics of the water tap 2 were characterized.

[0107] By introducing and As prior knowledge, the input tensor contains information about the physical causes of pressure curve fluctuations, enabling the model to distinguish between "normal rapid decay caused by an oversized opening" and "abnormal rapid decay caused by leakage".

[0108] S502, feature inference is performed based on a deep neural network model; the neural network prediction module 104 is internally loaded with a pre-trained deep neural network model. The model employs a dual-branch feature extraction architecture, the specific structure of which includes:

[0109] Temporal feature extraction branch: Employs Long Short-Term Memory (LSTM) or One-Dimensional Convolutional Neural Network (1D-CNN) layers specifically for processing... Extract the trend characteristics and local fluctuation characteristics of pressure decay;

[0110] Mechanistic feature mapping branch: Employs fully connected layers specifically for processing and This is mapped to a high-dimensional hidden layer state vector.

[0111] The outputs of the two branches are concatenated in the fusion layer and then subjected to nonlinear regression operations through a series of fully connected layers.

[0112] The model is initialized using a cold start strategy. Before the system leaves the factory, a dataset of standard samples (including qualified gold samples and silver samples with specific defects) covering different processing tolerance ranges is collected for offline pre-training to establish basic weight parameters. During actual operation, the model supports online fine-tuning and can iteratively update the weights based on the detection data that has been manually reviewed.

[0113] S503, generates dynamic acceptable pressure threshold range; neural network model. The output layer is configured as a regression output layer, containing two output nodes, which correspond to the minimum acceptable pressure value that the tested faucet 2 should have at the end of stage T2. and maximum qualified pressure value Output dynamic qualified pressure threshold range The computational model is expressed as follows:

[0114] ;

[0115] In the formula: This indicates the minimum pressure limit required to determine whether the product is qualified, taking into account the specific flow resistance and mechanical friction characteristics of the individual. This indicates the corresponding maximum pressure limit.

[0116] This threshold range is not a fixed value, but rather fluctuates dynamically depending on the physical state of each faucet 2. For example, for a faucet 2 identified as having "high flow resistance (small opening)" and "normal mechanical hysteresis," the model will automatically adjust the threshold range. and The value is determined because, physically, the smaller-opening faucet 2 should maintain a higher pipeline pressure; conversely, for the faucet 2 with "low flow resistance (larger opening)," the model will automatically lower the threshold range; the neural network prediction module 104 will calculate the dynamic qualified pressure threshold range. It is transmitted to the judgment execution module 105 as a dynamic benchmark for the final qualification judgment.

[0117] In the specific implementation of the decision execution module 105, in order to avoid the risk of misjudgment that may occur in artificial intelligence algorithms under extreme boundary conditions and to ensure the absolute reliability of the detection results at the physical mechanism level, this invention constructs a dual decision architecture based on parallel AI probabilistic prediction and mechanical deterministic verification; this architecture assigns the mechanical fault circuit interruption mechanism the highest logical priority relative to the AI ​​prediction results. The operation flow of the decision execution module 105 specifically includes the following sub-steps:

[0118] S601, Perform initial AI threshold compliance assessment; the judgment execution module 105 reads the measured pipeline pressure value at the end of the steady-state leakage stage T2, and records it as... Simultaneously, it receives the dynamic qualified pressure threshold range for the specific water tap 2 output by the neural network prediction module 104. .

[0119] The determination and execution module 105 will determine the measured pressure value. Compare with the dynamic acceptable pressure threshold range; if the measured pressure value meets the requirements... This indicates that the sealing performance and flow resistance characteristics of the water tap 2 numerically meet the prediction criteria of the AI ​​model. The execution module 105 then generates an AI pass signal, and this is recorded. Conversely, if the measured pressure value exceeds this range, an AI failure signal is generated and recorded. .

[0120] S602, execute the two-way mechanical fault circuit determination; the determination execution module 105 independently retrieves the mechanical hysteresis feature vector calculated by the feature extraction module 102 in the return verification stage T3. Hysteresis time difference The decision execution module 105 has a preset safe time range based on physical kinematic constraints. This range defines the physical boundary of the normal coupling action between the needle-shifting mechanism 10 and the valve core of the water nozzle 2; the determination module 105 determines the hysteresis time difference. A two-way boundary check is performed, and this check logic has the power to veto the final decision; the specific circuit breaker logic is as follows:

[0121] Scenario 1: High-level circuit breaker (anti-jamming); if This indicates that the valve core failed to reset promptly after the needle mechanism 10 was retracted, suggesting a physical fault such as mechanical jamming, spring fatigue, or residue buildup from cleaning. In this case, regardless of the AI ​​pass signal... If the value is 1, the execution module 105 will forcibly trigger the circuit breaker mechanism and generate a mechanical fault signal. ;

[0122] Scenario 2: Low-level fuse trips (to prevent loose connections); if This indicates that the pressure drop was detected in the initial reset phase of the needle-pulling mechanism 10, which means that the valve core may not have been effectively pushed open at all (e.g., the push rod stroke did not cover the valve core tolerance), or the valve core experienced a momentary false closure. Such situations would result in the pressure not effectively decreasing in the T2 stage, which could easily mislead the AI ​​model into judging it as "good sealing". In response to this situation, the judgment execution module 105 also forcibly triggers the fuse mechanism to generate a mechanical fault signal. .

[0123] Scenario 3: Normal passage; only when When this occurs, it indicates that the mechanical motion response of the valve core is within the normal physical range, and the execution module 105 generates a normal mechanical signal. .

[0124] S603, generates final control instructions and executes sorting; determines whether the AI ​​pass signal generated by S601 is valid by the execution module 105. The normal mechanical signal generated by S602 Perform a logical AND operation to obtain the final judgment result. The final decision-making logic formula is as follows:

[0125] ;

[0126] In the formula: This indicates the final sorting instruction; Represents the logical AND operator.

[0127] Based on the calculation results, if The central control module 100 determines that the current water tap 2 is a qualified product, and controls the servo conveyor belt 4 to continue running, so that the tooling fixture 7 carrying the water tap 2 passes through the rejection station and enters the finished product guide chute 15; if The central control module 100 determines that the current water nozzle 2 is a defective product. When the water nozzle 2 moves to the position of the rejection cylinder 14, the central control module 100 sends a drive signal to the rejection cylinder 14. The rejection cylinder 14 responds to the drive signal by extending the piston rod, pushing the defective water nozzle 2 out of the tooling fixture 7 and into the waste bin 11.

[0128] To further illustrate the application effect of the present invention in actual industrial production environments, especially the specific handling logic when facing gas source pressure fluctuation interference and hidden mechanical faults, a specific application scenario embodiment is provided below.

[0129] After the system is started, the scattered water nozzles 2 are automatically arranged by the vibratory plate and transported one by one to the fixed fixture 2 12 for pre-positioning in a set posture; when the water nozzles 2 reach the positioning point of the fixed fixture 2 12, the side air nozzles 13 are activated and spray high-pressure airflow to blow the surface of the water nozzles 2 and the water inlet to remove the cutting fluid or small iron filings remaining during the processing.

[0130] Subsequently, stepper motor 3 and stepper motor 5 work together to drive robot 6 to perform a transfer action; robot 6 grabs the cleaned water nozzle 2 from fixed fixture 12, transfers it and puts it into tooling fixture 7 on servo conveyor belt 4.

[0131] The above positioning, purging, gripping and loading actions are executed in a cycle until all four independent accommodating cavities on the tooling fixture 7 are fully loaded; the servo conveyor belt 4 is started to step-transport the tooling fixture 7, which is fully loaded with water nozzles 2, to the inspection station.

[0132] After the tooling fixture 7 arrives at the testing station, the fixing fixture 8 closes, pressing and fixing the tooling fixture 7 and the internal water nozzle 2 to prevent the workpiece from shifting due to subsequent testing pressure.

[0133] The sealing and air-injection mechanism 9 is pressed down vertically, and the sealing head at its end establishes an air circuit connection with the air inlet of the four water nozzles 2; the signal acquisition module 101 controls the air source valve to open, and fills the water nozzle with clean compressed test gas at 300 kPa (3.0 bar); the system performs a 100 ms static pressure holding pre-check: if the pipeline pressure cannot be maintained above 280 kPa before the opening is triggered, the system determines that the water nozzle at this station has a serious assembly leak, directly marks it as unqualified, and does not proceed to the subsequent dynamic test.

[0134] Suppose that one of the taps being inspected has a combined defect: machining burrs causing the valve core opening to be too small and foreign objects in the spring seat causing mechanical rebound lag; the system performs the inspection according to the following timing sequence:

[0135] During the dynamic opening phase T1 (0ms-200ms), the needle-pulling mechanism 10, driven by the motion control module 103, advances horizontally at a standard speed, contacting and pushing the valve core to produce a physical displacement of 0.8mm.

[0136] Because of the burrs on the faucet 2, the actual effective flow area formed by the valve core is smaller than the standard value; the feature extraction module 102 calculates the normalized flow resistance characteristic coefficient in real time. The average value was monitored. Significantly lower than the standard reference value (i.e., the flow resistance is too high).

[0137] Based on this, the system generates an opening fingerprint feature vector characterizing the "small opening / high flow resistance" property. .

[0138] Steady-state leakage stage T2 (200ms-800ms): The needle mechanism 10 remains locked in position, maintaining the valve core in the open state; gas continues to leak out of the pipeline, and the pressure decreases accordingly; due to the small opening of the water nozzle 2, the gas leakage speed is slow, resulting in residual pressure at the end of stage T2. It is 155 kPa (higher than the standard qualified product's usual 120 kPa).

[0139] The neural network prediction module 104 is activated; upon receiving... Based on the physical characteristics of the "small opening" transmitted, the model deduction suggests that for individuals with this type of flow resistance characteristics, the residual pressure of 155 kPa is consistent with the laws of fluid mechanics and is not an abnormal blockage.

[0140] Therefore, the model dynamically outputs the corrected acceptable pressure threshold range. is [150,160]kPa.

[0141] AI judgment result: The measured value of 155 kPa falls within the dynamic range, and the judgment execution module 105 generates an AI qualified signal. (If a traditional fixed threshold is used, the aerodynamic performance of this good product characteristic will be mistakenly judged as unqualified).

[0142] Return verification phase T3 (800ms-1000ms); Feedforward control execution: Motion control module 103 based on Based on feedback, the reset strategy is automatically adjusted to reduce the reset speed of the pin-pulling mechanism 10. Switch from the standard mode to a low-speed sensitive mode (e.g., 50 mm / s) to improve the sensitivity of detecting mechanical faults.

[0143] Hysteresis feature extraction: The needle-pulling mechanism 10 begins to retract; due to the presence of foreign objects in the spring seat, the valve core fails to return to its original position immediately after the needle-pulling mechanism, but instead experiences mechanical jamming, and only after a delay of 35ms does it spring back and close under the accumulation of spring force.

[0144] The feature extraction module 102 captures the inflection point of the pressure curve and calculates the hysteresis time difference. It takes 35ms.

[0145] Circuit breaker determination execution: The determination execution module 105 compares the preset physical safety range (e.g., [5,15]ms), confirms that 35ms is seriously exceeded, triggers the high-level circuit breaker mechanism, and generates a mechanical fault signal. .

[0146] At the end of the detection cycle, the judgment execution module 105 performs the final logical operation: The system ultimately determined that the water tap 2 was a defective product (fault type: mechanical jamming). The sealing and air injection mechanism 9 was lifted, and the servo conveyor belt 4 drove the tooling fixture 7 to move out of the inspection station.

[0147] When the tooling fixture 7 passes the position of the rejection cylinder 14, the central control module 100 drives the rejection cylinder 14 to extend according to the tracked non-conforming product station index, and accurately pushes the non-conforming water nozzle 2 into the waste bin 11.

[0148] The remaining qualified water taps 2 are transported to the end by the servo conveyor belt 4 and finally slide into the finished product guide trough 15 to complete the collection.

[0149] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A control system for detecting automatic drinking spouts in laboratory animals, applied to an automated detection device, characterized in that, The automated testing device includes a frame, a servo conveyor belt mounted on the frame, a testing station located on the conveying path of the servo conveyor belt, and a rejection cylinder located at the end of the servo conveyor belt; the servo conveyor belt conveys a tooling fixture that carries the water nozzle; the testing station is equipped with a first fixing fixture, a sealing gas injection mechanism, and a needle-pulling mechanism, the first fixing fixture clamps the water nozzle, the sealing gas injection mechanism injects test gas into the water nozzle, and the needle-pulling mechanism pushes the valve core of the water nozzle to actuate; The control system includes a central control module, which includes: The signal acquisition module is configured to acquire the absolute pressure signal of the sealing and gas injection mechanism and receive the motion status signal of the needle-pulling mechanism; The feature extraction module is configured to calculate the normalized flow resistance characteristic coefficient based on the absolute pressure signal and extract the mechanical hysteresis time difference feature during the reset process of the needle-pulling mechanism; wherein, the feature extraction module is configured to monitor the first derivative of the pressure data during the return verification phase of the needle-pulling mechanism performing the reset action; when it is detected that the pressure drop rate returns to zero or is lower than the preset background noise threshold, the moment when the pressure drop rate returns to zero or is lower than the preset background noise threshold is marked as the stop-fall inflection point moment; the feature extraction module calculates the time difference between the stop-fall inflection point moment and the start moment when the motion control module issues the reset command, and generates the mechanical hysteresis time difference feature; The motion control module is configured to generate motion commands for the needle-picking mechanism and adaptively adjust the reset speed of the needle-picking mechanism according to the normalized flow resistance characteristic coefficient. The neural network prediction module is configured to use a deep neural network model to output a dynamic qualified pressure threshold range based on the normalized flow resistance characteristic coefficient, mechanical hysteresis time difference characteristics, and pressure decay data. The judgment execution module is configured to control the rejection cylinder sorting based on the comparison results of the measured pressure value and the dynamic qualified pressure threshold range, as well as the verification results of the mechanical hysteresis time difference characteristics and the physical safety range.

2. The control system for detecting automatic drinking spouts in laboratory animals according to claim 1, characterized in that, The upper side of the frame is also provided with a stepper motor 1 and a stepper motor 2 located above the stepper motor 1. The stepper motor 2 is connected to a robotic arm. A fixing clamp 2 is provided on one side of the stepper motor 1, and an air blowing nozzle is provided on the side of the fixing clamp 2. The central control module is configured to control stepper motor one and stepper motor two to drive the robot to transfer the water nozzle to the fixed fixture two for pre-positioning, and to control the air nozzle to clean the water nozzle with airflow. Then, the robot is controlled to transfer the cleaned water nozzle to the tooling fixture on the servo conveyor belt.

3. The control system for detecting automatic drinking spouts in laboratory animals according to claim 1, characterized in that, The automated inspection device also includes a waste bin located next to the rejection cylinder and a finished product guide chute located at the end of the servo conveyor belt; the central control module is configured as follows: When the current water tap is determined to be a defective product, the rejection cylinder is controlled to push the water tap into the waste bin. When the current water tap is determined to be a qualified product, the servo conveyor belt is controlled to transport the water tap to the finished product guide trough.

4. The control system for detecting automatic drinking spouts in laboratory animals according to claim 1, characterized in that, The sealing and air-injection mechanism is configured to press down vertically to form a sealed connection with the air inlet end of the water nozzle; the needle-pulling mechanism is configured to move horizontally, and the needle-pulling mechanism includes a precision linear actuation unit equipped with a position feedback sensor; the signal acquisition module is configured to perform hardware clock synchronization calibration on the absolute pressure signal and the motion state signal of the needle-pulling mechanism to ensure that the timestamp of each pressure data point is strictly aligned with the timestamp of the position feedback data point.

5. The control system for detecting automatic drinking spouts in laboratory animals according to claim 1, characterized in that, The feature extraction module is configured to calculate the first derivative of the pressure data with respect to time during the dynamic opening phase when the needle-pulling mechanism pushes the valve core of the water nozzle to open, to obtain the instantaneous pressure drop rate, and define the ratio of the instantaneous pressure drop rate to the real-time absolute pressure value as the normalized flow resistance characteristic coefficient; the feature extraction module is also configured to statistically calculate the arithmetic mean, standard deviation and maximum value of the normalized flow resistance characteristic coefficient within the effective time window, and generate the opening fingerprint feature vector.

6. The control system for detecting automatic drinking spouts in laboratory animals according to claim 5, characterized in that, The motion control module is configured to read the arithmetic mean of the opening fingerprint feature vector before the needle-pulling mechanism performs the reset action; when the arithmetic mean is less than a preset flow resistance determination threshold, the motion control module determines that the current water nozzle has a large flow capacity and calls the speed parameter of the low-speed mode as the reset speed of the needle-pulling mechanism; when the arithmetic mean is greater than or equal to the flow resistance determination threshold, the speed parameter of the standard mode is called.

7. The control system for detecting automatic drinking spouts in laboratory animals according to claim 6, characterized in that, The feature extraction module further combines the mechanical hysteresis time difference feature with the current reset speed parameter to construct a mechanical hysteresis feature vector; the neural network prediction module is configured to obtain the pressure decay sequence of the steady-state leakage stage, the opening fingerprint feature vector generated in the dynamic opening stage, and the mechanical hysteresis feature vector generated in the back-trip verification stage; the neural network prediction module performs standardization processing and feature concatenation on the above data to construct a multidimensional input tensor.

8. The control system for detecting automatic drinking spouts in laboratory animals according to claim 7, characterized in that, The deep neural network model loaded in the neural network prediction module includes a temporal feature extraction branch and a mechanism feature mapping branch; The output layer of the deep neural network model is configured as a regression output layer, used to output the dynamic qualified pressure threshold range containing the minimum qualified pressure value and the maximum qualified pressure value, the dynamic qualified pressure threshold range dynamically fluctuating with the change of the input multidimensional input tensor.

9. The control system for detecting automatic drinking spouts in laboratory animals according to claim 8, characterized in that, The decision execution module is configured to execute dual circuit breaker decision logic: The first level of judgment is the artificial intelligence compliance judgment. When the measured pressure value at the end of the steady-state leakage stage falls within the dynamic qualified pressure threshold range, an artificial intelligence qualified signal is generated. The second level of determination is mechanical fault determination. When the mechanical hysteresis time difference characteristic exceeds the preset physical safety range, a mechanical fault signal is generated. The physical safety range is defined by the high-level time threshold for anti-jamming and the low-level time threshold for anti-loose connection. The determination and execution module is configured to perform logical operations on the artificial intelligence qualified signal and the mechanical fault signal; if and only if the artificial intelligence qualified signal is generated and the mechanical fault signal is not generated, the current water tap is determined to be a qualified product; if the mechanical fault signal is generated, regardless of whether the artificial intelligence qualified signal exists, the current water tap is forcibly determined to be a defective product and a rejection instruction is generated.

Images