Oil seal assembly process intelligent regulation and control method and system fused with time sequence state perception

Abstract

The invention provides an oil seal assembly process intelligent regulation and control method and system fused with time sequence state perception, and relates to the technical field of intelligent control. The method comprises the steps that a multi-mode sensing array is rigidly integrated on press-fitting equipment; at the moment of press-fitting starting, press-fitting process data acquisition with the aligned timestamps is synchronously executed, and multi-mode associated data streams associated with time sequences are output; executing state online feature extraction, and generating a dynamic multi-modal feature vector; carrying out press-fitting process hidden state prediction, and outputting a real-time press-fitting decision instruction; the press-fitting equipment is controlled to execute step-by-step press-fitting operation; and according to the time-varying characteristics of the multi-mode associated data flow, press-fitting control closed-loop regulation and control are executed until the oil seal is press-fitted in place. The technical problems that in the prior art, due to the fact that a traditional sensor is generally dependent on monitoring of a single physical quantity in the press-fitting process, complex and multi-layer physical changes are difficult to capture, the oil seal press-fitting quality is difficult to control accurately, and the final sealing effect is affected are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent control, in particular to an oil seal assembly process intelligent regulation method and system fusing time sequence state perception. BACKGROUND

[0002] The oil seal assembly plays a crucial role in the production process of engines or other mechanical equipment, especially the installation of the sealing lip. The sealing lip and the outer edge of the oil seal are usually made of rubber materials with strong elasticity and viscoelasticity. If improper operation occurs during the installation of the oil seal, problems such as distortion, aging, and failure may occur due to excessive compression, insufficient relaxation, or uneven distribution of lubricants, which not only affect the sealing performance of the oil seal but also affect the long-term reliability of the equipment, and even cause serious oil leakage.

[0003] The existing technology usually relies on traditional sensors to monitor a single physical quantity during the pressing process, which is difficult to capture complex and multi-level physical changes, resulting in ineffective control of key factors such as stress relaxation and lubricant distribution during the pressing process, and further leading to difficulties in accurately controlling the oil seal pressing quality, problems such as excessive stress and uneven distribution of lubricants, and affecting the final sealing effect. SUMMARY

[0004] In view of the above defects or improvement needs of the prior art, the present application provides an oil seal assembly process intelligent regulation method and system fusing time sequence state perception, which is used to solve the technical problem that the existing technology usually relies on traditional sensors to monitor a single physical quantity during the pressing process, which is difficult to capture complex and multi-level physical changes, resulting in difficulties in accurately controlling the oil seal pressing quality, and affecting the final sealing effect.

[0005] In a first aspect, the present application provides an oil seal assembly process intelligent regulation method fusing time sequence state perception, which comprises:

[0006] rigidly integrating a multi-modal sensing array on a pressing device; synchronously activating the multi-modal sensing array to perform time stamp alignment of the pressing process data acquisition at the moment of starting the pressing of the pressing device, outputting time sequence correlated multi-modal correlation data stream; performing online state feature extraction on the multi-modal correlation data stream to generate dynamic multi-modal feature vector; performing pressing process hidden state prediction based on the dynamic multi-modal feature vector to output real-time pressing decision instruction; controlling the pressing device to perform step-by-step pressing operation according to the real-time pressing decision instruction; and performing pressing control closed-loop regulation according to the time-varying characteristics of the multi-modal correlation data stream until the oil seal is pressed into place.

[0007] In an embodiment, the dynamic multi-modal feature vector is used to perform hidden state prediction of the press-fitting process, and output real-time press-fitting decision instructions, and the following processing is further performed:

[0008] A discrete hidden state set is defined based on the physical stages of the oil seal press-fitting, wherein the discrete hidden state set includes an initial compression state, a stress relaxation critical state, a lubricant distribution steady state, and a press-fitting to position final state. State transition probability and observation probability parameters of a hidden Markov model are configured based on the discrete hidden state set, and model parameter loading is completed. The dynamic multi-modal feature vector is loaded into the hidden Markov model as an observation sequence, and the maximum a posteriori probability state is calculated in real time based on a forward algorithm. The real-time press-fitting decision instructions are output according to the cross-stage state transition attribute of the maximum a posteriori probability state.

[0009] In an embodiment, state online feature extraction is performed on the multi-modal associated data stream to generate a dynamic multi-modal feature vector, and the following processing is further performed:

[0010] Multi-thread online feature extraction is performed on the multi-modal associated data stream to obtain a pressure decay curve, a displacement rate curve, and a lubricant coverage area change rate curve. Dynamic feature extraction is performed on the pressure decay curve, the displacement rate curve, and the lubricant coverage area change rate curve mapping to obtain a pressure decay slope sequence, a displacement turning point marker sequence, and a lubricant coverage area change rate sequence. The pressure decay slope sequence, the displacement turning point marker sequence, and the lubricant coverage area change rate sequence are assembled and integrated based on a time axis to output the dynamic multi-modal feature vector.

[0011] In an embodiment, multi-thread online feature extraction is performed on the multi-modal associated data stream to obtain a pressure decay curve, a displacement rate curve, and a lubricant coverage area change rate curve, and the following processing is further performed:

[0012] The pressure signal stream, the displacement signal stream, and the visual image stream in the multi-modal associated data stream are respectively input into a sliding decay fitting container, a difference calculation container, and a dynamic ROI tracking container. Real-time exponential decay fitting is performed on the pressure signal stream in the sliding decay fitting container to output the pressure decay curve. First-order difference operation is performed on the displacement signal stream in the difference calculation container to generate an original instantaneous speed sequence, and then sliding average filtering is performed on the original instantaneous speed sequence to output the displacement rate curve. After the visual image stream is segmented in the HSV space in the dynamic ROI tracking container, adjacent frame area change rate calculation is performed based on the segmentation result to output the lubricant coverage area change rate curve. The sliding decay fitting container, the difference calculation container, and the dynamic ROI tracking container perform multi-thread parallel feature extraction.

[0013] In an embodiment, the following processing is further performed:

[0014] The maximum a posteriori probability state is parsed to output the real-time state type, state duration, and state transition direction. If the real-time state type of the maximum a posteriori probability state is the initial compression state and the state duration is ≥2 frames, then the timing of the compression process is triggered.

[0015] In one implementation, if the real-time state type of the maximum a posteriori probability state is a stress relaxation critical state and the state duration is ≥3 frames, then the first step pressing end point command is output.

[0016] In one implementation, if the state transition direction of the maximum a posteriori probability state is from the stress relaxation critical state to the lubricant distribution steady state, then a dynamic pause duration command is output.

[0017] In one implementation, if the real-time state type of the maximum a posteriori probability state is a lubricant distribution steady state and the state duration is ≥5 frames, then the second step pressing speed and termination pressure command are output.

[0018] In one embodiment, the multimodal sensor array comprises an axial pressure sensor integrated into the load-bearing structure of the press-fitting fixture, an axial displacement sensor integrated into the base of the press-fitting equipment, and a lip vision sensor fixed to the anti-vibration bracket, wherein the anti-vibration bracket is decoupled from the press-fitting equipment.

[0019] A second aspect of this invention discloses an intelligent control system for oil packaging and assembly processes that integrates time-series state perception. The system is used in the aforementioned intelligent control method for oil packaging and assembly processes that integrates time-series state perception. The system includes:

[0020] The system includes a sensor array integration unit for rigidly integrating a multimodal sensor array into the pressing equipment; a data acquisition unit for synchronously activating the multimodal sensor array to perform timestamp-aligned pressing process data acquisition at the moment the pressing equipment starts, and outputting a time-series correlated multimodal data stream; a feature extraction unit for performing online feature extraction on the multimodal data stream to generate a dynamic multimodal feature vector; a latent state prediction unit for predicting the latent state of the pressing process based on the dynamic multimodal feature vector, and outputting a real-time pressing decision command; a pressing operation execution unit for controlling the pressing equipment to perform step-by-step pressing operations according to the real-time pressing decision command; and a closed-loop control unit for performing closed-loop control of the pressing operation according to the time-varying characteristics of the multimodal data stream until the oil seal is pressed in place.

[0021] One or more technical solutions provided in this invention have at least the following technical effects or advantages:

[0022] The rigid integrated multi-modal sensor array is integrated in the press-fitting device, and by synchronously activating the sensors, multi-dimensional real-time data such as axial pressure, displacement, visual image and the like can be acquired simultaneously, the data stream of each sensor has a clear time stamp, thereby ensuring that the timing data in the press-fitting process is highly synchronized, and accurate state perception is provided; by performing online feature extraction on the multi-modal associated data stream, effective features can be extracted in real time from multiple dimensions, and the generation of dynamic multi-modal feature vectors enables the quantification of various important features in the press-fitting process, thereby providing accurate basis for subsequent hidden state prediction and decision instruction generation, and improving the accuracy and response speed of intelligent control; by associating the dynamic multi-modal feature vector with the hidden state in the press-fitting process, the hidden state of the current press-fitting process can be predicted in real time, and the prediction can effectively identify the key turning points in the oil seal press-fitting process, thereby providing a basis for the next step of press-fitting control, and the generated real-time press-fitting decision instruction includes specific adjustment of the press-fitting process, thereby ensuring accurate control of the press-fitting process; according to the real-time decision instruction, the press-fitting device can perform accurate step-by-step press-fitting operation, thereby ensuring accurate control of the press-fitting depth and improving the press-fitting quality; in the press-fitting process, real-time feedback is provided to the system according to the time-varying characteristics of the multi-modal associated data stream, closed-loop control is supported, and by continuously adjusting the press-fitting strategy until the oil seal is completely press-fitted in place, the assembly quality and precision of the oil seal are ensured. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0024] Figure 1 A fusion time sequence state perception oil seal assembly process intelligent regulation and control method flowchart provided by the present application is shown.

[0025] Figure 2 A fusion time sequence state perception oil seal assembly process intelligent regulation and control system structure diagram provided by the present application is shown.

[0026] Legend: sensor array integration unit 1, data acquisition unit 2, feature extraction unit 3, hidden state prediction unit 4, press-fitting operation execution unit 5, closed-loop regulation and control unit 6. DETAILED DESCRIPTION

[0027] The application provides a fusion timing state-aware oil seal assembly process intelligent regulation method and system, and aims to solve the technical problem that the prior art usually relies on traditional sensors to monitor a single physical quantity in the press-fitting process, it is difficult to capture complex and multi-level physical changes, the oil seal press-fitting quality is difficult to accurately control, and the final sealing effect is affected.

[0028] In order to make the object, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and not to limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0029] In the description of the present application, it should be understood that the orientations or positional relationships indicated by the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", etc. are based on the orientations or positional relationships shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application.

[0030] Unless otherwise explicitly stated, throughout the specification and claims, the term "comprise" or its variants such as "comprises" or "comprising" will be understood to encompass the stated elements or components, without excluding other elements or components.

[0031] In one embodiment, the present application provides a fusion timing state-aware oil seal assembly process intelligent regulation method, referring to Figure 1 , the method comprises:

[0032] The rigid integration of the multi-modal sensor array is performed on the press-fitting equipment.

[0033] In the oil seal assembly process, in order to accurately monitor various physical changes in the assembly process, multiple types of sensors need to be integrated, including: an axial pressure sensor for real-time measurement of the press-fitting force applied to the oil seal; an axial displacement sensor for measuring the displacement of the oil seal during the press-fitting process; a lip visual sensor for monitoring the contact between the oil seal lip and the shaft, capturing problems such as distortion and overturning. All these sensors are rigidly integrated into the press-fitting equipment to ensure stability and accuracy during data acquisition. Rigid integration means that the sensor integration must have sufficient rigidity to avoid unnecessary deformation during the press-fitting process, thereby affecting the accuracy of the data.

[0034] At the moment of the press-fitting starting of the press-fitting device, the multi-modal sensor array is synchronously activated to perform time-stamp aligned press-fitting process data acquisition, and output time-series correlated multi-modal correlated data stream.

[0035] The moment of the press-fitting starting refers to the moment when the press-fitting device starts to apply pressure to the oil seal. At this moment, all sensors are synchronously activated to ensure that they start collecting data at the same time. This synchronization ensures that all sensor data is aligned based on the same time point. Each sensor records the collected data, including pressure, displacement, visual image, etc., according to the precise time stamp. The time stamp allows all sensor data to be accurately aligned in subsequent analysis, ensuring that multi-modal data at each moment can be reasonably correlated. Ultimately, all real-time data from different sensors are combined into a multi-modal correlated data stream, which is a multi-modal data stream with time stamps. This data stream contains multiple data types, all of which have time stamps and can accurately reflect the state and changes of the press-fitting process.

[0036] Performing state online feature extraction on the multi-modal correlated data stream generates a dynamic multi-modal feature vector.

[0037] State online feature extraction refers to extracting useful information from the multi-modal correlated data stream for subsequent analysis. Here, feature extraction is real-time, i.e., online. This means that as the data stream is collected, the algorithm begins to process the data and extract key features such as pressure decay curve, displacement rate curve, and lubricant distribution. The extracted features are organized into a dynamic multi-modal feature vector, which is a composite feature set that can fully reflect the key physical states in the press-fitting process. It not only contains data from a single sensor, but also integrates data from multiple sensors to provide more comprehensive and accurate process monitoring.

[0038] Based on the dynamic multi-modal feature vector, the press-fitting process hidden state is predicted, and real-time press-fitting decision instructions are output.

[0039] The hidden state in the press-fitting process refers to states that cannot be directly identified from sensors, including initial compression state, stress relaxation critical state, lubricant distribution steady state, and press-fitting to position final state. These hidden states need to be inferred through existing data patterns and statistical methods. Here, a hidden Markov model is used to predict the hidden state. The model takes the dynamic feature vector as input and calculates the hidden state at each moment. Through the prediction of the hidden state, the next operation of the press-fitting process is determined, and real-time press-fitting decision instructions are output. For example, if the "stress relaxation critical state" is predicted, the press-fitting device is instructed to pause for 2 seconds for stress relaxation; if the "lubricant distribution steady state" is entered, the press-fitting device is instructed to perform accurate press-fitting to ensure accurate oil seal position.

[0040] The real-time press-fitting decision instruction controls the press-fitting equipment to perform a step-by-step press-fitting operation.

[0041] From the start to the end of the press-fitting, the entire process needs to be divided into multiple stages, such as a preliminary press-fitting stage, a stress relaxation stage, and a fine press-fitting stage. Each stage has different requirements and different press-fitting speeds, pressures, and times. The real-time press-fitting decision instruction indicates how the press-fitting equipment adjusts. For example, in the preliminary press-fitting stage, only the oil seal needs to be quickly pressed into a certain position, while in the fine press-fitting stage, a more accurate press-fitting depth is required. In different stages, the press-fitting speed of the equipment needs to be adjusted, for example, in the stress relaxation stage, the press-fitting speed needs to be slowed down to avoid rapidly compressing the rubber and generating excessive internal stress. By embedding the real-time press-fitting decision instruction into the control algorithm of the press-fitting equipment, the press-fitting process is adjusted in real time to ensure that each stage is performed according to the predetermined standard and to provide correct feedback for subsequent operations.

[0042] According to the time-varying characteristics of the multi-modal associated data stream, a press-fitting control closed-loop regulation is performed until the oil seal is press-fitted in place.

[0043] As the press-fitting process progresses, the multi-modal associated data stream changes, and these changes are constantly monitored and responded to according to the time-varying characteristics of the data stream. For example, the pressure rises rapidly in the initial stage and then stabilizes, and the displacement speed slows down as the oil seal gradually enters the groove. The press-fitting control closed-loop regulation is automatically adjusted according to real-time feedback, which means that if the pressure or displacement in a certain stage does not change as expected, the action of the press-fitting equipment is immediately adjusted, for example, if it is found that the lubricant is not evenly distributed, the press-fitting is paused and the lubricant is redistributed. The closed-loop regulation is continuously performed until it is ensured that the oil seal is completely press-fitted in place and no abnormalities occur. Through the monitoring of time-varying data, fine regulation can be performed at each link to ensure that the press-fitting precision and quality are optimal.

[0044] Further, based on the dynamic multi-modal feature vector, a press-fitting process hidden state prediction is performed, and a real-time press-fitting decision instruction is output, the method comprising:

[0045] A discrete hidden state set is defined based on the physical stages of the oil seal press-fitting, wherein the discrete hidden state set includes an initial compression state, a stress relaxation critical state, a lubricant distribution stable state, and a press-fitting in place final state. The state transition probability and observation probability parameters of the hidden Markov model are configured based on the discrete hidden state set, and the model parameter loading is completed. The dynamic multi-modal feature vector is loaded as an observation sequence into the hidden Markov model, and the maximum a posteriori probability state is calculated in real time based on the forward algorithm. According to the cross-stage state transition attribute of the maximum a posteriori probability state, the real-time press-fitting decision instruction is output.

[0046] The discrete hidden state set is based on the physical stages in the oil seal press-fitting process, which decomposes the press-fitting process into multiple key hidden states. These hidden states reflect different physical manifestations of the oil seal during the assembly process, which cannot be directly observed by sensors, but can be inferred by indirect features. Among them, the initial compression state is usually accompanied by a rapid rise in pressure; the stress relaxation critical state refers to the process of the rubber material undergoing stress relaxation, i.e., the molecular chain begins to rearrange, and the internal stress gradually releases. This stage is characterized by a slowdown in pressure increase and stable displacement; the lubricant distribution steady state refers to the uniform distribution of lubricant, ensuring that the oil seal can smoothly contact the shaft and form a seal. This state marks that the lubricant of the oil seal has been evenly distributed, and the press-fitting process enters a stable state; the press-fitting to the final state refers to the oil seal being completely pressed into the predetermined installation position, and the press-fitting process ends. The characteristics are that the changes of pressure and displacement tend to be stable, reaching the termination state.

[0047] The defined discrete hidden state set is converted into a hidden Markov model, which infers the current hidden state from the observed sensor data and makes the next decision based on the predicted hidden state. The state transition probability describes the probability of transitioning from one hidden state to another, such as the transition probability from the initial compression state to the stress relaxation critical state, or the transition probability from the lubricant distribution steady state to the press-fitting to the final state. These transition probabilities are estimated based on historical data and physical laws, reflecting the natural transition between different physical stages. The observation probability describes the probability of observing feature data through sensors given a hidden state, such as observing higher pressure and slower displacement change in the initial compression state. There is a certain correlation between these observation data and hidden states. Through historical data, the observation probability of each hidden state can be calculated by the maximum likelihood estimation method. The state transition probability and observation probability parameters are loaded into the hidden Markov model to establish a complete model, which provides a basis for subsequent hidden state prediction.

[0048] The dynamic multi-modal feature vector is input into the hidden Markov model as an observation sequence, each feature vector corresponding to a time point in the press-fitting process, describing the sensor data at that time. The forward algorithm is an efficient algorithm for calculating the probability of each hidden state in the hidden Markov model given an observation sequence. Through recursive calculation, the forward algorithm can calculate the most likely hidden state at each time point. This algorithm updates the probability distribution of the hidden state step by step from the initial time to the current time, and finally calculates the maximum a posteriori probability state, which represents the current most likely physical state.

[0049] By analyzing the transition between the current hidden state and the hidden state at the previous moment, it is determined which stage the current is in and whether a state jump is needed, for example, from the stress relaxation critical state to the lubricant distribution steady state, which represents that the press-fitting process has entered a new stage. According to the predicted state transition, the corresponding real-time press-fitting decision instruction is output to adjust the action of the press-fitting equipment, for example, if it is predicted that the stress relaxation critical state will jump to the lubricant distribution steady state, the equipment is instructed to pause and wait for the uniform distribution of the lubricant; if the state transition indicates that the oil seal has reached the press-fitting-in-place terminal state, the instruction to stop press-fitting is issued.

[0050] Further, the method comprises:

[0051] The multi-threaded online feature extraction is performed on the multi-modal associated data stream to obtain a pressure decay curve, a displacement rate curve, and a lubricant coverage area change rate curve; dynamic feature extraction is performed on the pressure decay curve, the displacement rate curve, and the lubricant coverage area change rate curve mapping to obtain a pressure decay slope sequence, a displacement turning point marker sequence, and a lubricant coverage area change rate sequence; and the pressure decay slope sequence, the displacement turning point marker sequence, and the lubricant coverage area change rate sequence are assembled and integrated based on a time axis to output the dynamic multi-modal feature vector.

[0052] The multi-threaded parallel processing processes the signal streams from different sensors, and each thread processes one data stream to extract the key features of the data stream, wherein the pressure decay curve represents the curve of the change of pressure with time during the press-fitting process, which usually presents the characteristics of rapid rise at the beginning and subsequent decay; the displacement rate curve represents the press-fitting speed of the oil seal, i.e., the displacement change rate, which reflects the change of the displacement speed of the oil seal during the press-fitting process; and the lubricant coverage area change rate curve represents the change of the lubricant coverage on the oil seal lip, which reflects how the lubricant spreads and distributes from both sides of the sealing lip as the press-fitting process progresses.

[0053] Further dynamic feature extraction is performed from the extracted curves. Dynamic feature extraction refers to extracting key features from the original curves that can reflect state changes. Specifically, during the press-fitting process, the pressure decay curve usually exhibits a process of initial rapid rise and subsequent slow decline. By calculating the slope of the pressure decay curve, a pressure decay slope sequence is obtained. The slope reflects the rate of pressure change, which affects the press-fitting quality of the oil seal during the press-fitting process. Changes in the slope sequence can be used to predict whether ideal press-fitting conditions have been reached. The displacement rate curve can exhibit a linear change, followed by a clear turning point, which represents certain key moments in the press-fitting process, such as stress relaxation or changes in lubricant distribution. The turning point marker sequence records the positions of important turning points in the displacement curve, which usually reflect changes in the contact state of the oil seal during press-fitting. The lubricant coverage area change rate curve records the distribution of lubricant on the oil seal lip. By performing dynamic feature extraction on this curve, a lubricant coverage area change rate sequence is obtained, which reflects the uniformity and stability of lubricant distribution and helps determine whether the lubricant has been completely distributed on the oil seal lip, thereby reducing local hydraulic impact.

[0054] Each feature sequence has a corresponding timestamp, reflecting its time change during the press-fitting process. The integration process aligns the three sequences along the time axis to ensure that they reflect the timing relationship during the press-fitting process. After integrating the three sequences along the time axis, a dynamic multi-modal feature vector containing all key information is generated. This vector not only contains the data of each feature sequence but also contains their timing features during the press-fitting process.

[0055] Further, multi-threaded online feature extraction is performed on the multi-modal associated data stream to obtain the pressure decay curve, displacement rate curve, and lubricant coverage area change rate curve. The method includes:

[0056] The pressure signal stream, displacement signal stream, and visual image stream in the multi-modal associated data stream are respectively input into a sliding decay fitting container, a difference calculation container, and a dynamic ROI tracking container. The sliding decay fitting container performs real-time exponential decay fitting on the pressure signal stream, outputting the pressure decay curve. The difference calculation container performs first-order difference operation on the displacement signal stream, generating an original instantaneous speed sequence, and then performs sliding average filtering on the original instantaneous speed sequence, outputting the displacement rate curve. The dynamic ROI tracking container performs HSV space segmentation on the visual image stream, and then calculates the area change rate of adjacent frames based on the segmentation results, outputting the lubricant coverage area change rate curve. The sliding decay fitting container, difference calculation container, and dynamic ROI tracking container perform multi-threaded parallel feature extraction.

[0057] The pressure signal stream is a data stream from the pressure sensor, reflecting the change of pressure in the press-fitting process, which is input into the sliding attenuation fitting container for processing the pressure signal and obtaining the pressure attenuation curve; the displacement signal stream is a data stream from the displacement sensor, describing the change of displacement of the oil seal in the press-fitting process, which is input into the difference calculation container for difference operation to calculate the displacement rate and further for smoothing processing; the visual image stream is a data stream from the visual sensor, reflecting the distribution of the oil seal and the lubricant, which is input into the dynamic ROI tracking container for tracking the distribution of the lubricant on the oil seal lip, and further obtaining the lubricant coverage area change rate curve.

[0058] The function of the sliding attenuation fitting container is to perform exponential attenuation fitting on the pressure signal stream. The pressure signal usually presents a trend of initial rapid rise and then gradual attenuation in the press-fitting process, which can be fitted by an exponential function. After the pressure signal stream is input, the sliding attenuation fitting container calculates the change trend of the pressure in real time, and applies an exponential attenuation model to fit the data. Exponential attenuation shows that the rising speed of the pressure gradually slows down after a certain time point and tends to be stable. Through fitting, a smooth pressure attenuation curve is obtained, reflecting the dynamic characteristics of the change of pressure in the press-fitting process.

[0059] The main task of the difference calculation container is to perform first-order difference operation on the displacement signal stream, aiming to calculate the instantaneous displacement rate of the oil seal in the press-fitting process. Since displacement is a function of time, the speed change can be obtained by performing difference operation on the displacement signal. First-order difference calculation on the displacement signal obtains the original instantaneous speed sequence, reflecting the displacement change rate of the oil seal at each time step, that is, the instantaneous speed in the press-fitting process. The change rate of displacement can reflect the press-fitting state of the oil seal. For example, when the oil seal gradually contacts the shaft, the press-fitting speed is slow, and when the press-fitting reaches a certain critical point, the press-fitting speed may change. Smoothing average filtering is performed on the original instantaneous speed sequence to smooth out the noise in the data, making the displacement rate curve smoother and more stable. The sliding average is a filtering technique that reduces sudden noise interference by calculating the average value of data in a time window. The filtered data is the displacement rate curve, which reflects the smooth displacement rate in the press-fitting process of the oil seal, helping to judge whether the press-fitting is smoothly carried out.

[0060] The dynamic ROI tracking container is responsible for extracting the lubricant coverage area change rate from the visual image stream and outputting the lubricant coverage area change rate curve. HSV space is a color space that includes hue (Hue), saturation (Saturation), and value (Value) three components, which is more suitable for color segmentation than RGB (Red Green Blue) space, especially under different lighting conditions, the color recognition will be more stable. By performing HSV space segmentation on the visual image stream, the lubricant area can be effectively extracted from the image, because the lubricant usually has a specific color and texture, and this segmentation operation classifies the pixel values in the image by color to find the lubricant area.

[0061] Adjacent frames refer to two consecutive images in time. By comparing the changes in the lubricant area in these adjacent frames, the coverage area change of the lubricant in this time period can be obtained. The area change rate of the lubricant coverage area in adjacent frames, that is, the change amount of the lubricant coverage area per unit time, is calculated to monitor whether the distribution of the lubricant is uniform, for example, if the lubricant is not evenly distributed during the press fitting process, it may cause poor sealing or excessive local pressure. The output is the lubricant coverage area change rate curve, which can reflect the change of the lubricant over time during the press fitting process.

[0062] In order to ensure that the above data processing can be performed in real time, during feature extraction, multi-thread parallel processing is adopted, each container is executed in parallel on different threads and does not block each other, thereby improving the overall processing efficiency.

[0063] Further, the method further comprises:

[0064] The maximum a posteriori state is analyzed for attributes, and the real-time state type, state duration, and state transition direction are output; if the real-time state type of the maximum a posteriori state is an initial compressed state, and the state duration is greater than or equal to 2 frames, the press fitting process start timing is triggered.

[0065] The maximum a posteriori state is the most likely state calculated by the hidden Markov model, which represents the hidden state of the current press fitting process at a specific time point. The maximum a posteriori state is analyzed for attributes, the real-time state type is the physical phase of the current press fitting process, the state duration represents the time that the current state lasts at this time, and the state transition direction refers to the transition from the current state to the next state.

[0066] The real-time state type is the initial compression state, indicating that the oil seal is undergoing initial compression, at this time the rubber is in a relatively large stress state, starting to contact the shaft and bear the compression force, the pressing equipment starts to exert pressure on the oil seal, but it has not yet entered the complete installation state. The state duration refers to the time from the initial compression state to the current time, that is, the time during which the oil seal remains in this state, based on 2 frames, meaning that the oil seal has been in this state for at least 2 frames, which is usually about 0.067 seconds at a certain frame rate, for example, 30 frames per second. In this state, it indicates that the oil seal has entered the initial compression stage and the pressing process has started, triggering the start of the pressing process timing, providing a time base for subsequent operations of the pressing process.

[0067] Further, if the real-time state type of the maximum a posteriori probability state is the stress relaxation critical state and the state duration is ≥ 3 frames, output the first step pressing end instruction.

[0068] The real-time state type is the stress relaxation critical state, indicating that during the oil seal pressing process, the rubber has entered the stress relaxation stage, in which the molecular chains of the rubber are rearranging and the pressure is starting to release. This is a transitional state, indicating that the pressing process of the oil seal is changing, the rubber is no longer experiencing large stress, but is gradually relaxing. The duration of the stress relaxation critical state is monitored, and if this state has been maintained for at least 3 frames, it indicates that the stress of the rubber has gradually relaxed and the process is close to completing the initial pressing. The 3 frames are a set threshold to ensure that the stress relaxation of the rubber is sufficient and the system can prepare for the next step of the pressing operation. In this case, the first step pressing end instruction is output, instructing the pressing equipment to stop the current stage of pressing action and prepare for the next step of operation.

[0069] Further, if the state transition direction of the maximum a posteriori probability state is the stress relaxation critical state transitioning to the lubricant distribution stable state, output the dynamic pause duration instruction.

[0070] The state transition direction is the stress relaxation critical state transitioning to the lubricant distribution stable state, indicating that the oil seal has experienced stress relaxation of the rubber, at this time the distribution of the lubricant between the oil seal and the shaft gradually tends to be stable, and the lubricant is no longer concentrated in a certain area but is uniformly distributed. At this time, the dynamic pause duration instruction is output, instructing the pressing equipment to temporarily stop pressing. The purpose of this pause instruction is to ensure that the lubricant has sufficient time to uniformly exit from the sealing lip area of the oil seal and be uniformly distributed on the entire contact surface. This process requires a certain amount of time to avoid high pressure caused by the concentration of lubricant and ensure a more stable subsequent pressing process.

[0071] Further, if the real-time state type of the maximum a posteriori probability state is the lubricant distribution stable state and the state duration is ≥ 5 frames, output the second step pressing speed and termination pressure instruction.

[0072] The real-time state type is lubricant distribution steady state, indicating that the lubricant between the oil seal and the shaft has been evenly distributed and is no longer affected by the hydraulic lock or uneven lubrication. The lubricant distribution steady state ensures smooth progress of the press-fitting process and prevents local high pressure caused by accumulation of lubricant, which may cause damage to the oil seal. The duration of the lubricant distribution steady state is continuously monitored. If the uniform distribution of lubricant between the oil seal and the shaft has lasted for at least 5 frames, it indicates that the lubricant has been sufficiently distributed, and the second step of precise press-fitting can be started. The 5 frames is a preset time threshold, indicating that the lubricant distribution has reached a sufficiently steady state, which is sufficient to ensure that the subsequent press-fitting process will not be affected by uneven lubrication. In this case, the second step press-fitting speed and termination pressure instructions are output, and the second step press-fitting operation is started. The second step press-fitting usually requires precise speed control to ensure that the oil seal will not be subjected to excessive pressure during the final press-fitting process, thereby affecting the sealing effect. The termination pressure instruction marks the completion of the final press-fitting, ensuring that the oil seal will not continue to be excessively pressed when it reaches the predetermined press-fitting depth.

[0073] Further, the multi-modal sensing array includes an axial pressure sensor integrated into the load-bearing structure of the press-fitting clamp, an axial displacement sensor integrated into the base of the press-fitting device, and a lip visual sensor fixed to a vibration isolation support, wherein the vibration isolation support is decoupled from the press-fitting device.

[0074] The axial pressure sensor is used to monitor the pressure experienced by the oil seal during the press-fitting process in real time. The pressure sensor is installed in the load-bearing structure of the press-fitting clamp and directly interacts with the oil seal and its contact surface, enabling it to sense changes in the load borne by the rubber during the press-fitting process. When the press-fitting clamp applies pressure, the sensor records real-time changes in axial pressure. These changes reflect the mechanical state during the press-fitting process, such as stress concentration, excessive press-fitting, or normal press-fitting process, etc.

[0075] The axial displacement sensor is responsible for monitoring the axial displacement of the oil seal during the press-fitting process. It is installed in the base of the press-fitting device and can accurately measure the press-in depth and displacement rate of the oil seal. The axial displacement sensor obtains real-time displacement data by detecting the axial movement of the oil seal during the press-fitting process. These data are used to determine whether the oil seal has been correctly positioned and whether there is a risk of incomplete press-fitting.

[0076] The lip visual sensor is used to capture the contact state between the oil seal lip and the shaft. It is installed on a vibration isolation support that is decoupled from the press-fitting device to reduce the impact of vibration on the sensor and ensure its stability and accuracy. The visual sensor captures the contact area between the oil seal and the shaft and analyzes the deformation of the lip, the distribution of lubricant, and the contact condition of the sealing lip using image processing technology, thereby providing visual feedback for subsequent press-fitting operations.

[0077] Embodiment two, based on the same inventive concept as the intelligent regulation method for oil seal assembly process integrating timing state awareness in the preceding embodiment, as shown in the figure, the embodiment of the present application provides an intelligent regulation system for oil seal assembly process integrating timing state awareness, which comprises: Figure 2

[0078] a sensor array integration unit 1 for rigidly integrating a multi-modal sensor array in a press-fitting device; a data acquisition unit 2 for synchronously activating the multi-modal sensor array to perform time stamp aligned press-fitting process data acquisition at the moment of press-fitting start of the press-fitting device, outputting time sequence correlated multi-modal correlated data stream; a feature extraction unit 3 for performing state online feature extraction on the multi-modal correlated data stream, generating dynamic multi-modal feature vector; a hidden state prediction unit 4 for performing press-fitting process hidden state prediction based on the dynamic multi-modal feature vector, outputting real-time press-fitting decision instruction; a press-fitting operation execution unit 5 for controlling the press-fitting device to perform step-by-step press-fitting operation according to the real-time press-fitting decision instruction; a closed-loop regulation unit 6 for performing press-fitting control closed-loop regulation according to the time-varying features of the multi-modal correlated data stream until the oil seal is press-fitted in place.

[0079] Further, the hidden state prediction unit 4 is configured to perform the following operation steps:

[0080] defining a discrete hidden state set based on the physical stages of oil seal press-fitting, wherein the discrete hidden state set comprises initial compression state, stress relaxation critical state, lubricant distribution steady state, and press-fitting in place terminal state; configuring the state transition probability and observation probability parameters of the hidden Markov model based on the discrete hidden state set, completing model parameter loading; loading the dynamic multi-modal feature vector as an observation sequence to the hidden Markov model, and calculating the maximum a posteriori probability state in real time based on the forward algorithm; outputting the real-time press-fitting decision instruction according to the cross-stage state transition attribute of the maximum a posteriori probability state.

[0081] Further, the feature extraction unit 3 is configured to perform the following operation steps:

[0082] performing multi-thread online feature extraction on the multi-modal correlated data stream to obtain pressure decay curve, displacement rate curve, and lubricant coverage area change rate curve; performing dynamic feature extraction on the pressure decay curve, displacement rate curve, and lubricant coverage area change rate curve mapping to obtain pressure decay slope sequence, displacement turning point marker sequence, and lubricant coverage area change rate sequence; assembling and integrating the pressure decay slope sequence, displacement turning point marker sequence, and lubricant coverage area change rate sequence based on the time axis to output the dynamic multi-modal feature vector.

[0083] Further, the feature extraction unit 3 is configured to perform the following operation steps:​

[0084] The pressure signal stream, displacement signal stream and visual image stream in the multi-modal correlation data stream are input into a sliding attenuation fitting container, a difference calculation container and a dynamic ROI tracking container respectively; the sliding attenuation fitting container performs real-time exponential attenuation fitting on the pressure signal stream, and outputs the pressure attenuation curve; the difference calculation container performs first-order difference operation on the displacement signal stream, generates an original instantaneous speed sequence, and then performs sliding average filtering on the original instantaneous speed sequence to output the displacement rate curve; the dynamic ROI tracking container performs HSV space segmentation on the visual image stream, and based on the segmentation result, calculates the area change rate of adjacent frames to output the lubricant coverage area change rate curve; wherein the sliding attenuation fitting container, difference calculation container and dynamic ROI tracking container perform multi-thread parallel feature extraction.

[0085] Further, the hidden state prediction unit 4 is configured to perform the following operation steps:

[0086] The maximum a posteriori state is subjected to attribute analysis, and the real-time state type, state duration and state transition direction are output; if the real-time state type of the maximum a posteriori state is an initial compression state, and the state duration is greater than or equal to 2 frames, the press fitting process starting timing is triggered.

[0087] Further, if the real-time state type of the maximum a posteriori state is a stress relaxation critical state, and the state duration is greater than or equal to 3 frames, a first step press fitting end instruction is output.

[0088] Further, if the state transition direction of the maximum a posteriori state is a stress relaxation critical state to a lubricant distribution steady state, a dynamic pause duration instruction is output.

[0089] Further, if the real-time state type of the maximum a posteriori state is a lubricant distribution steady state, and the state duration is greater than or equal to 5 frames, a second step press fitting speed and termination pressure instruction is output.

[0090] Further, the multi-modal sensing array includes an axial pressure sensor integrated in a press fitting clamp force bearing structure, an axial displacement sensor integrated in a press fitting device base, and a lip visual sensor fixed to a vibration isolation support, wherein the vibration isolation support is decoupled from the press fitting device.

[0091] The above only describes the preferred embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for intelligent regulation of oil seal assembly process with fusion of timing state awareness, characterized by, include: Rigidly integrated multimodal sensor array in press-fitting equipment; At the instant the pressing equipment starts pressing, the multimodal sensor array is simultaneously activated to perform timestamp-aligned pressing process data acquisition and output a time-series correlated multimodal data stream; Perform online state feature extraction on the multimodal associated data stream to generate a dynamic multimodal feature vector; Based on the dynamic multimodal feature vector, the hidden state of the pressing process is predicted, and real-time pressing decision instructions are output. Based on the real-time pressing decision command, the pressing equipment is controlled to perform step-by-step pressing operations; Based on the time-varying characteristics of the multimodal correlated data stream, closed-loop control of the pressing process is performed until the oil seal is pressed into place.

2. The intelligent control method for oil packaging and assembly process integrating time-series state perception as described in claim 1, characterized in that, Based on the dynamic multimodal feature vector, the hidden state of the pressing process is predicted, and real-time pressing decision instructions are output, including: Based on the physical stages of oil seal press-fitting, a discrete hidden state set is defined, which includes the initial compression state, the stress relaxation critical state, the lubricant distribution steady state, and the final state of press-fitting in place. Based on the discrete hidden state set, configure the state transition probability and observation probability parameters of the hidden Markov model to complete the model parameter loading; The dynamic multimodal feature vector is loaded into the hidden Markov model as an observation sequence, and the maximum posterior probability state is calculated in real time based on the forward algorithm. Based on the cross-stage state transition attributes of the maximum a posteriori probability state, the real-time pressing decision instruction is output.

3. The intelligent control method for oil packaging and assembly process integrating time-series state perception as described in claim 1, characterized in that, Perform online state feature extraction on the multimodal correlated data stream to generate dynamic multimodal feature vectors, including: Multi-threaded online feature extraction is performed on the multimodal correlated data stream to obtain pressure decay curve, displacement rate curve and lubricant coverage area change rate curve; Dynamic feature extraction is performed on the pressure decay curve, displacement rate curve, and lubricant coverage area change rate curve to obtain a pressure decay slope sequence, a displacement inflection point marker sequence, and a lubricant coverage area change rate sequence. Based on the time axis, the pressure decay slope sequence, displacement inflection point marker sequence, and lubricant coverage area change rate sequence are assembled and integrated to output the dynamic multimodal feature vector.

4. The intelligent control method for oil packaging and assembly process integrating time-series state perception as described in claim 3, characterized in that, Multi-threaded online feature extraction is performed on the multimodal correlated data stream to obtain pressure decay curves, displacement rate curves, and lubricant coverage area change rate curves, including: The pressure signal stream, displacement signal stream, and visual image stream in the multimodal correlated data stream are respectively input into the sliding attenuation fitting container, the difference calculation container, and the dynamic ROI tracking container; The pressure signal stream is fitted with real-time exponential decay in the sliding decay fitting container, and the pressure decay curve is output. After the differential calculation container performs a first-order differential operation on the displacement signal stream to generate the original instantaneous velocity sequence, the displacement rate curve is output by performing a moving average filter on the original instantaneous velocity sequence. After the dynamic ROI tracking container performs HSV spatial segmentation on the visual image stream, it calculates the area change rate of adjacent frames based on the segmentation results and outputs the lubricant coverage area change rate curve. The sliding decay fitting container, the difference calculation container, and the dynamic ROI tracking container perform multi-threaded parallel feature extraction.

5. The intelligent control method for oil packaging and assembly process integrating time-series state perception as described in claim 2, characterized in that, Also includes: The maximum a posteriori probability state is analyzed for attributes, and the real-time state type, state duration and state transition direction are output. If the real-time state type of the maximum a posteriori probability state is the initial compression state, and the state duration is ≥2 frames, then the timing of the compression process is triggered.

6. The intelligent control method for oil packaging and assembly process integrating time-series state perception as described in claim 5, characterized in that, If the real-time state type of the maximum a posteriori probability state is a stress relaxation critical state, and the state duration is ≥3 frames, then the first step pressing end point command is output.

7. The intelligent control method for oil packaging and assembly process integrating time-series state perception as described in claim 5, characterized in that, If the state transition direction of the maximum a posteriori probability state is from the stress relaxation critical state to the lubricant distribution steady state, then output a dynamic pause duration command.

8. The intelligent control method for oil packaging and assembly process integrating time-series state perception as described in claim 5, characterized in that, If the real-time state type of the maximum a posteriori probability state is a lubricant distribution steady state and the state duration is ≥5 frames, then output the second step pressing speed and termination pressure command.

9. The intelligent control method for oil packaging and assembly process integrating time-series state perception as described in claim 1, characterized in that, The multimodal sensor array comprises an axial pressure sensor integrated into the load-bearing structure of the press-fitting fixture, an axial displacement sensor integrated into the base of the press-fitting equipment, and a lip vision sensor fixed to the anti-vibration bracket, wherein the anti-vibration bracket is decoupled from the press-fitting equipment.

10. An intelligent control system for oil packaging and assembly process integrating time-series state perception, characterized in that, The system is used to implement the intelligent control method for oil packaging and assembly process with fusion of time-series state perception as described in any one of claims 1-9, the system comprising: A sensor array integration unit is used to rigidly integrate a multimodal sensor array into a press-fitting device; The data acquisition unit is used to synchronously activate the multimodal sensor array to perform timestamp-aligned data acquisition of the pressing process at the moment the pressing equipment starts pressing, and output a time-series correlated multimodal data stream; The feature extraction unit is used to perform online state feature extraction on the multimodal associated data stream and generate a dynamic multimodal feature vector; The hidden state prediction unit is used to predict the hidden state of the pressing process based on the dynamic multimodal feature vector and output real-time pressing decision instructions. The pressing operation execution unit is used to control the pressing equipment to perform step-by-step pressing operations according to the real-time pressing decision command; The closed-loop control unit is used to perform closed-loop control of the pressing operation based on the time-varying characteristics of the multimodal associated data stream until the oil seal is pressed into place.

Images