Intelligent grinding device for floating oil seal and online closed-loop regulation process

Abstract

The present application relates to the technical field of mechanical seal processing equipment and processing technology, and specifically relates to a floating oil seal intelligent grinding device and an online closed-loop regulation process. The device comprises a grinding device body, which comprises a rack, a grinding disc, at least one limiting tool, a grinding execution mechanism and a driving mechanism. The grinding disc is installed on the rack. The limiting tool is arranged inside the grinding disc. The grinding execution mechanism is arranged above the limiting tool. The driving mechanism is installed at the bottom of the rack. A multi-source sensing and collecting component is integrally arranged on the grinding device body. A quality online detection system is in communication connection with the multi-source sensing and collecting component, and is used for online identification of workpiece processing quality according to multi-dimensional process parameters. An intelligent regulation system is connected with the quality online detection system, the driving mechanism and the multi-source sensing and collecting component, and is used for self-adaptive adjustment of grinding process parameters according to the quality identification result.

Description

Technical Field

[0001] This invention relates to the field of mechanical seal processing equipment and processing technology, specifically to a floating oil seal intelligent grinding device and online closed-loop control process. Background Technology

[0002] A floating oil seal is a special mechanical seal device designed for low-speed, heavy-load conditions. Its basic structure consists of a pair of metal sealing rings and a pair of rubber sealing rings. It typically needs to be used in conjunction with the equipment's sealing cavity (floating seal seat) to form a complete sealing system. The metal sealing rings are usually made of wear-resistant alloy materials such as high-chromium cast iron or chromium-molybdenum alloy cast iron, through casting and precision machining, achieving a hardness of HRC 60 or higher. The rubber sealing rings are made of oil-resistant and aging-resistant elastomer materials such as nitrile rubber (NBR), hydrogenated nitrile rubber (H-NBR), or fluororubber (FKM). During assembly, the rubber rings are installed on the back of the metal rings, and both are inserted into the floating seal seat. The elastic deformation of the rubber rings generates axial preload, ensuring that the precision-ground end faces of the pair of metal rings remain tightly fitted, forming a reliable seal during relative rotational motion.

[0003] When a floating oil seal is in operation, an extremely thin lubricating film forms between the relatively rotating end faces of the two metal rings. This lubricating film simultaneously performs two core functions: sealing and lubrication. On the one hand, it fills the microscopic gap between the end faces, relying on fluid dynamic pressure and viscous shear force to prevent lubricating oil leakage and the intrusion of external impurities. On the other hand, it completely separates the two metal end faces, transforming solid-solid friction into fluid friction, significantly reducing the coefficient of friction and frictional heat. The thickness and stability of the lubricating film directly determine the sealing effect and service life of the floating oil seal, and the lubricating film is highly dependent on the surface roughness, flatness, parallelism, and dimensional accuracy of the sealing end faces. Related studies have shown that the initial sealing surface width of the floating seal is only 2mm, and the eccentricity must be controlled within 0.2mm to ensure long-term stable operation of the equipment.

[0004] From a manufacturing perspective, the stability of the lubricating film is highly dependent on the machining precision of the metal sealing ring, especially key quality parameters such as surface roughness, flatness, parallelism, and dimensional accuracy of the sealing end face. Studies show that the initial sealing surface width of a floating oil seal is only 2mm, and long-term reliable operation can only be guaranteed when the eccentricity does not exceed 10% (i.e., 0.2mm). However, current floating oil seal machining technology has fundamental shortcomings in the following aspects: (I) Equipment Level: Insufficient Structural Rigidity and Lack of Online Sensing Capabilities. Most current floating oil seal grinding equipment is a purely mechanical open-loop structure, resulting in insufficient overall rigidity and susceptibility to deformation in transmission and pressure transmission components. Process parameters such as grinding pressure and grinding disc speed rely on manual adjustment, leading to poor adjustment accuracy and slow response. Furthermore, the equipment lacks sensor acquisition modules, making traditional equipment a "black box" processing mode. This makes it impossible to trace process problems, establish standardized process systems, and collect real-time data on physical quantities containing process status information, such as machine tool spindle vibration, grinding disc pressure, processing area temperature, and acoustic emission signals. The processing process remains in a "black box" state.

[0005] (II) Inspection Level: Quality inspection relies on offline sampling, and the quality status is unknown. Currently, the quality inspection of floating oil seals mainly relies on offline sampling or full inspection after processing, using equipment such as coordinate measuring machines, surface roughness testers, and optical flats to perform post-processing measurements on the sealing rings. This mode is not only inefficient and costly, but also has two inherent defects: first, it cannot avoid the generation of defective products, and by the time a problem is discovered, a large number of products may already be defective; second, the inspection results cannot be fed back to the processing in real time, making it impossible to make timely process adjustments to the current workpiece.

[0006] (III) Control Level: Process control relies on manual experience, making precise quality control impossible. Due to the lack of process monitoring data, the current control of floating oil seal processing quality is highly dependent on manual experience—when quality fluctuations occur, operators adjust process parameters manually based on visual observation, touch, or even hearing. This control method varies from person to person and from time to time, lacking unified operating standards and scientific basis. At the same time, processing disturbances such as abrasive passivation, pressure fluctuations, and temperature changes cannot be detected and compensated for in a timely manner, further leading to low product qualification rates and poor quality consistency.

[0007] As construction machinery continues to develop towards larger scale, heavier load capacity, and greater intelligence, the main equipment places higher demands on the reliability and durability of floating oil seals. Therefore, developing an intelligent grinding device for floating oil seals with online sensing, online detection, and intelligent adaptive control capabilities, along with its supporting processes, has become a pressing technical problem to be solved in this field. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of existing floating oil seal grinding equipment, such as insufficient rigidity, low parameter adjustment accuracy, lack of online sensing capability, lagging quality inspection, reliance on manual experience for process control, poor product consistency, low pass rate, and difficulty in quality traceability. This invention provides a floating oil seal intelligent grinding device and an online closed-loop control process.

[0009] The technical solution adopted in this invention is: a floating oil seal intelligent grinding system, comprising: a grinding device body, the grinding device body including a frame, a grinding disc, at least one limiting fixture, a grinding execution mechanism, and a driving mechanism; the grinding disc is mounted on the frame for placing workpieces; the limiting fixture is located inside the grinding disc; the grinding execution mechanism is arranged above the limiting fixture; the driving mechanism is mounted on the bottom of the frame for driving the grinding disc to rotate; a multi-source sensing acquisition component, integrated on the grinding device body, for real-time acquisition of multi-dimensional process parameters during the grinding process; an online quality detection system, communicatively connected to the multi-source sensing acquisition component, for online identification of workpiece processing quality based on the multi-dimensional process parameters; and an intelligent control system, connected to the online quality detection system, the driving mechanism, and the multi-source sensing acquisition component, for adaptively adjusting the grinding process parameters based on the quality identification results.

[0010] The present invention is further configured such that: the multi-source sensing acquisition component includes a pressure sensor, a vibration sensor, a temperature sensor, an acoustic emission sensor, a speed sensor, and an energy consumption acquisition module, used to simultaneously acquire multi-dimensional process parameters such as grinding pressure, equipment vibration, processing temperature, acoustic emission signal, grinding disc speed, and processing energy consumption; the multi-source sensing acquisition component is laid with shielded cables and establishes a communication connection with the online quality detection system through an industrial Ethernet.

[0011] The present invention is further configured such that: the online quality detection system includes a data preprocessing subsystem, a feature extraction subsystem, and a quality identification subsystem; the data preprocessing subsystem uses wavelet transform to denoise the signal, combines moving average filtering to smooth time-series data, and uses the 3σ criterion to remove outliers; the feature extraction subsystem extracts time-domain, frequency-domain, and time-frequency-domain features of the data, and uses correlation analysis to select core feature parameters related to processing quality, constructing a quality feature dataset; the quality identification subsystem uses a CNN+LSTM fusion network model, with the quality feature dataset as training samples; the CNN+LSTM fusion network model sequentially passes through a convolutional neural network, feature reconstruction, a long short-term memory network, and a fully connected unit to complete local spatial feature extraction, temporal dimension transformation, temporal feature mining, and quality result output, respectively.

[0012] The present invention is further configured such that: the quality identification subsystem is deployed on an edge computing terminal to realize local real-time computation of feature extraction and quality determination.

[0013] The present invention is further configured such that: the intelligent control system includes a knowledge graph module, a parameter mapping module, and an adaptive execution module; the knowledge graph module stores processing knowledge using a triplet structure of "quality problem - fault cause - key parameter - control scheme" and uses a deterministic factor algorithm to complete fault logic reasoning; the parameter mapping module uses a multiple linear regression algorithm to establish a quantitative correlation between process parameters and workpiece quality indicators; the adaptive execution module issues control commands to the drive mechanism based on the quality identification results and the calculation results of the parameter mapping module, and automatically adjusts the grinding process parameters.

[0014] The present invention is further configured such that: the grinding device body also includes a pressure regulating mechanism, which is disposed above the grinding execution mechanism and is used to regulate the grinding pressure applied to the workpiece by the grinding execution mechanism; the pressure regulating mechanism includes an electric cylinder and an electrically controlled adjusting nut, and the driving mechanism is a servo motor.

[0015] A closed-loop control process for intelligent grinding of floating oil seals includes the following steps: S1: Clamp the floating oil seal workpiece in the limiting fixture and start the entire system; S2: Multi-source sensor acquisition components collect multi-dimensional process parameters in real time during the grinding process; S3: The online quality inspection system preprocesses and extracts features from the collected data to identify the current processing quality status of the workpiece; S4: If the processing quality deviates, the intelligent control system matches the corresponding control strategy through the knowledge graph and calculates the target adjustment amount of the process parameters through the parameter mapping model. S5: The intelligent control system sends control commands to the drive mechanism to adjust the grinding process parameters in real time, and the equipment does not need to be stopped during the parameter adjustment process. S6: Repeat steps S2 to S5 to form a closed loop of the entire process of "data acquisition - quality identification - parameter adjustment - reprocessing" until the grinding process is completed.

[0016] The present invention is further configured such that, in step S3, the preprocessing includes: using wavelet transform to achieve signal denoising, using moving average filtering to achieve data smoothing, and using the 3σ criterion to remove abnormal data.

[0017] The present invention is further configured such that: in step S3, the feature extraction includes: extracting time domain, frequency domain, and time-frequency domain features, and using the Pearson correlation coefficient analysis method to screen core feature parameters related to processing quality.

[0018] The present invention is further configured such that: the quality deviation mentioned in step S4 includes at least one of the following: excessive surface roughness of the sealing end face, non-compliance with flatness, non-compliance with parallelism, dimensional deviation, abnormal vibration caused by abrasive passivation, and abnormal processing temperature; the closed-loop response speed of the intelligent control system from quality identification to parameter adjustment is controlled within 100ms.

[0019] Compared with the prior art, the present invention has the following advantages: I. Through structural optimization and electrification upgrades, significantly improve equipment processing accuracy and condition awareness capabilities. This invention is based on fully parametric 3D virtual modeling and multiphysics coupling simulation. It accurately locates the weak areas of structural rigidity through static simulation, analyzes vibration errors and motion deviations under different working conditions through dynamic simulation, and iteratively optimizes the pressure transmission structure, limit tooling structure and transmission mechanism clearance of the grinding actuator. It solves the problems of insufficient rigidity and low processing accuracy of traditional equipment from the design source.

[0020] Meanwhile, this invention replaces the traditional manual adjustment structure with an electric servo pressure adjustment mechanism that combines an electric cylinder and an electric control adjustment nut, and replaces the drive mechanism with an electric servo drive component, realizing digital continuous and precise adjustment of key process parameters such as grinding pressure and grinding disc speed, and greatly improving parameter response speed and control accuracy.

[0021] Based on this, the present invention integrates six types of sensing units, namely pressure sensor, vibration sensor, temperature sensor, acoustic emission sensor, speed sensor and energy consumption acquisition module, in the grinding actuator, pressure adjustment mechanism and spindle position. Shielded cables are used for laying and high-speed data transmission is achieved through industrial Ethernet, enabling the equipment to have the ability to collect physical quantities of the entire processing process in real time, which completely changes the "unsensory" technical status of traditional floating oil seal grinding equipment.

[0022] II. Construct a multi-source integrated online quality inspection system to achieve real-time monitoring of processing quality. This invention develops a dedicated multi-sensor synchronous data acquisition subsystem, configured with a high sampling rate data acquisition terminal, which realizes high-speed, synchronous, and distortion-free acquisition of multi-dimensional process parameters such as grinding pressure, equipment vibration, processing temperature, acoustic emission signal, grinding disc speed, and processing energy consumption. The acquisition frequency is precisely matched with the grinding process cycle, capturing full processing status information.

[0023] In terms of data processing, this invention employs wavelet transform to denoise the original sensor signals, uses moving average filtering to smooth the time-series data, and uses the 3σ criterion to remove outliers, effectively eliminating the impact of environmental noise and equipment interference on data analysis. Based on this, multi-dimensional features in the time domain, frequency domain, and time-frequency domain are extracted, and core key parameters strongly correlated with processing quality are accurately identified through Pearson correlation coefficient analysis. This constructs a dedicated quality feature dataset, achieving scientific dimensionality reduction from massive amounts of raw data to effective features.

[0024] In terms of quality identification, this invention innovatively designs a CNN+LSTM fusion deep learning model. It extracts local spatial features through a convolutional neural network, reconstructs these features, and then mines temporal correlation features using a long short-term memory network. Finally, a fully connected layer outputs the quality judgment result, enabling real-time and accurate judgment of key quality indicators such as surface roughness, flatness, parallelism, and dimensional accuracy of the floating oil seal sealing end face. The judgment result is synchronized with the processing, completely replacing the traditional offline sampling inspection mode. The model is deployed on an edge computing terminal, realizing local real-time computation of feature extraction and quality judgment, ensuring that the detection response speed matches the grinding process cycle. Simultaneously, a visual monitoring interface is provided, automatically triggering an alert when quality anomalies are detected, achieving visualized and traceable management of the processing quality status.

[0025] Third, establish a data- and knowledge-driven intelligent control mechanism to achieve precise and controllable processing quality. This invention constructs a process knowledge graph module with a ternary structure of "quality problem - failure cause - key parameter - control scheme", which transforms the scattered knowledge that traditionally relies on human experience into a structured, digital, and visualized computable knowledge base, solving the technical problem that traditional experience is difficult to pass on and cannot be standardized.

[0026] This invention integrates real-time process data from a quality inspection system with expert experience from a knowledge graph. It employs a deterministic factor algorithm to complete fault logic reasoning and a multiple linear regression algorithm to establish a quantitative mapping model between process parameters and workpiece quality indicators. This constructs a dual-engine control mechanism that combines data-driven and knowledge-driven approaches, transforming process control from "fuzzy judgment" relying on human experience to "quantitative calculation" based on scientific evidence.

[0027] Based on this, the present invention has developed an adaptive execution module. When the quality identification result deviates from the qualified range or there is a risk of exceeding the tolerance, the system automatically matches the control strategy and calculates the optimal process parameter adjustment amount. By issuing control commands to the electric servo pressure adjustment mechanism and the drive mechanism, the system realizes the adaptive and precise adjustment of process parameters such as grinding pressure and grinding disc speed. The adjustment process does not require the equipment to be stopped. The system completes the full closed-loop control from quality identification to parameter adjustment with a response speed of less than 100ms, realizing dynamic real-time compensation for processing disturbances.

[0028] IV. Construct a three-in-one, fully closed-loop collaborative system to achieve deep integration of hardware, perception, and decision-making. This invention overcomes the drawbacks of fragmented processes in traditional processing technologies by constructing a fully closed-loop intelligent processing system with the synergistic effect of three core technologies: an improved grinding device, a multi-source online quality detection system, and a data-knowledge fusion intelligent control system. The improved grinding device provides the hardware carrier and sensor integration platform for the detection system, and a precise execution terminal for the control algorithm. The quality detection system, through real-time data acquisition and analysis, connects hardware devices and intelligent algorithms, providing real-time and reliable status input for control. The intelligent control system, based on detection data and expert knowledge, reverse-optimizes the processing parameters of the grinding device, forming a closed-loop process of "equipment processing → data acquisition → quality identification → parameter control → equipment reprocessing → quality verification." The deep integration and synergistic effect of these three technologies systematically solves the core technical challenges of floating oil seal grinding processing from the three dimensions of equipment, detection, and control, achieving an overall upgrade of the processing technology.

[0029] V. Significantly improve processing quality and production efficiency to meet the stringent requirements of high-end engineering machinery. By optimizing equipment precision, implementing real-time process monitoring, and accurately controlling parameters, this invention effectively improves the control accuracy of key quality indicators such as surface roughness, flatness, parallelism, and dimensional accuracy of the floating oil seal sealing end face. Comparative experiments have verified that, after adopting the device and process of this invention, the eccentricity of the floating oil seal sealing end face is stably controlled from 0.22mm-0.45mm with traditional equipment to 0.10mm-0.18mm, meeting the long-term reliable operation requirement of eccentricity not exceeding 0.2mm; the product qualification rate has increased from 82% with traditional equipment to over 98%, significantly improving the stability, consistency, and qualification rate of processing quality.

[0030] Meanwhile, real-time online quality identification and proactive closed-loop control enable early prediction, timely warning, and precise adjustment of quality issues, avoiding the generation of defective products during processing, significantly reducing subsequent rework and scrap costs, and improving production efficiency. Improved processing quality directly optimizes the contact state and lubrication film formation quality of the floating oil seal's sealing end face, significantly enhancing the sealing performance and service life of the floating oil seal. This fully meets the stringent requirements of high-end engineering machinery such as excavators, tunneling machines, and continuous coal mining machines for core sealing components under low-speed, heavy-load conditions, while reducing downtime losses and maintenance costs caused by seal failure. This invention is adaptable to the processing of floating oil seals of different specifications and materials. The algorithm and model can be continuously iterated and optimized based on historical data. Standardized hardware interfaces facilitate functional expansion, demonstrating promising engineering application prospects and large-scale promotion value.

[0031] The embodiments of the present invention will be further described below with reference to the accompanying drawings. Attached Figure Description

[0032] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a structural diagram of the grinding machine in this invention; Figure 2 This is a schematic diagram of the floating oil seal textured surface lubrication model in this invention; Figure 3 This is a flowchart of the data acquisition process based on multiple sensors in this invention; Figure 4 This invention relates to knowledge graph-driven processing quality analysis and control. Figure 5 This is a schematic diagram illustrating the principle of lubricating oil film pressure formation in the floating oil seal of this invention. Detailed Implementation

[0033] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are for illustrative purposes only and should not be considered as limiting the scope of the invention. Simple modifications and equivalent substitutions made by those skilled in the art based on the technical solution of the present invention are all within the protection scope of the present invention.

[0034] Specific structure and assembly of the floating oil seal intelligent grinding device: like Figure 1 - Figure 5 As shown, the main body of this device is a vertical frame 1. A worktable is set at the bottom of the frame 1, and a customized annular limiting fixture 8 is installed at the center of the worktable. This limiting fixture is an annular limiting iron ring. The workpiece is placed inside the annular limiting iron ring. The fixture can be changed according to the floating oil seal with different outer diameters to prevent the workpiece from shifting during the grinding process. A drive mechanism is assembled at the top of the frame 1. The drive mechanism consists of a servo motor and a transmission wheel. The transmission wheel is connected to a vertically arranged grinding spindle. The lower end of the spindle is fixed with a grinding disc 6 and a grinding pad 7.

[0035] The electric servo pressure adjustment mechanism is located next to the spindle, employing an electric cylinder 2 paired with electrically controlled adjusting nuts 3 and 4. This eliminates the need for a traditional manual screw structure, allowing for continuous and precise adjustment of the grinding clamping force via electrical control signals. Piezoelectric pressure sensors, triaxial vibration sensors, contact temperature sensors, acoustic emission sensors, and speed sensors are integrated into the cylinder sidewall, grinding spindle, and grinding disc base, respectively. The entire machine is equipped with an energy consumption acquisition module. All sensors utilize shielded cable wiring and interact with the backend system via industrial Ethernet.

[0036] The multi-source online quality inspection system and the data-knowledge fusion intelligent control system are integrated into the main control cabinet of the equipment, with the hardware deployed on the edge computing terminal to ensure the real-time performance of data processing and model calculation. After the equipment is assembled, assembly verification is carried out based on the results of the previous 3D modeling and multiphysics simulation. No-load testing, load testing, full sensor calibration and standard workpiece trial cutting are carried out in sequence to complete the overall performance calibration of the machine.

[0037] After the device is assembled, no-load and load tests are carried out in sequence to calibrate the sensor acquisition accuracy and servo mechanism adjustment accuracy. Then, a trial cutting experiment is conducted using a standard floating oil seal workpiece to verify the overall processing performance.

[0038] In use, the floating oil seal metal sealing ring workpiece to be processed is placed at the center position of the limiting fixture 8, and the annular limiting iron ring restricts the large displacement of the workpiece. The cylinder 2 is activated, applying grinding pressure to the grinding disc 6 through the pressure plate 5. The grinding pad 7 contacts the upper surface of the workpiece, and the drive mechanism drives the grinding disc 6 to rotate, grinding the end face of the workpiece. The grinding pressure can be precisely adjusted by the upper and lower electrically controlled adjusting nuts 3 and 4, and the grinding pressure and speed can be adjusted in real time by electric servo control.

[0039] Simulation-based device structure optimization implementation method: First, based on the mechanical structure, dimensional parameters and actual processing conditions of existing floating oil seal grinding equipment, a fully parameterized high-precision three-dimensional virtual model is established to completely restore the structure and assembly relationship of core components such as grinding head, worktable, drive mechanism, and pressure adjustment components.

[0040] Secondly, multi-physics coupling simulation analysis is carried out: by static simulation, the stress distribution and deformation characteristics of the core components during the grinding process are simulated to identify the weak links in the structural rigidity, including the bending deformation area of ​​the pressure transmission structure of the grinding head and the insufficient rigidity area of ​​the worktable support structure; by dynamic simulation analysis, the vibration characteristics and motion accuracy error of the equipment under different grinding pressures and speeds are analyzed to locate the key structural defects that affect the flatness and parallelism of the machining.

[0041] Finally, based on the simulation results, iterative optimization design was carried out: the pressure transmission structure of the grinding head was optimized to solve the problem of insufficient rigidity, the worktable limit fixture 8 was optimized to improve the positioning accuracy, and the clearance of the transmission mechanism was optimized to improve the motion accuracy, so as to ensure that the structural rigidity and motion positioning accuracy of the improved equipment meet the processing requirements of high-end floating oil seals.

[0042] The multi-source sensing and acquisition component includes a piezoelectric pressure sensor, a triaxial vibration sensor, a contact temperature sensor, an acoustic emission sensor, a speed sensor, and an energy consumption acquisition module. Each sensor connects to a high-sampling-rate data acquisition terminal via a standardized interface.

[0043] After the device is started, the multi-source sensing acquisition component synchronously collects raw data at a sampling frequency matching the grinding process (a sampling frequency of no less than 10 kHz is adopted in this embodiment), and uploads the data to the multi-sensor synchronous acquisition subsystem via the TCP / IP protocol. The data acquisition subsystem adopts a hardware acquisition and data transmission module based on a front-end and back-end separation architecture: the front-end completes real-time acquisition and preliminary conversion of multi-source sensing signals, and the back-end realizes high-speed transmission of acquired data through industrial Ethernet. The system is equipped with a time-series database to classify and store the acquired data in real time, supporting fast data retrieval, traceback and visualization.

[0044] The acquired data is classified and stored in real time using a time-series database, supporting fast data retrieval, traceback and visual display. Aiming at the problem of environmental interference in on-site machining, a shielding layer is added in the sensing circuit design for hardware anti-interference, a verification mechanism is added in data transmission, and a filtering module is developed at the software level to ensure the integrity and reliability of the acquired data.

[0045] For problems such as environmental noise, equipment interference and outliers existing in the collected raw multi-source data, the data preprocessing subsystem sequentially adopts three methods for processing: wavelet denoising, moving average filtering and 3σ criterion outlier elimination.

[0046] (1) Wavelet Denoising Wavelet transform is used to denoise the original sensing signal. The basic principle of wavelet transform is to decompose the signal into wavelet coefficients of different scales and positions by performing inner product operation on the signal and the mother wavelet after scale stretching and time translation, so as to realize the separation of signal and noise in the time-frequency domain. The calculation formula of wavelet transform is:

[0047] In the formula, is the wavelet transform coefficient of the signal at scale a and translation position b, which reflects the local characteristics of the signal at this time-frequency point; the scale factor a is used to control the stretching degree of the mother wavelet, where a>1 corresponds to the low-frequency slow-varying characteristics of the signal, and 0<a<1 corresponds to the high-frequency fast-varying characteristics; the translation factor b is used to control the position of the mother wavelet on the time axis, so as to realize the time positioning of signal characteristics; is the mother wavelet function that meets the conditions of zero mean and limited energy is its complex conjugate (for real signals ); t is the time variable, is the normalization factor, which is used to ensure the energy consistency of wavelets at different scales; the integral operation essentially calculates the inner product of the signal and the mother wavelet after scale stretching and time translation, so as to measure the local similarity between the two, thereby realizing the separation of noise and effective features in the signal.

[0048] In this embodiment, the Daubechies wavelet is used as the mother wavelet, and the decomposition level is 5. The formula for calculating the approximation coefficients of the wavelet decomposition is as follows:

[0049] In the formula, For the signal via the first The approximate coefficient sequence obtained after layer wavelet decomposition is used to characterize the low-frequency backbone information of the signal; The wavelet decomposition level is denoted as . This is the index of the coefficients for the current layer; These are the coefficients of the low-pass filter. This serves as the index for the coefficient sequence of the previous level. For the first The approximation coefficients obtained from layer wavelet decomposition are then convolved with a low-pass filter and downsampled to obtain the approximation coefficients for this layer. .

[0050] By using wavelet decomposition, the high-frequency noise components in the signal are separated into detail coefficients. After thresholding the detail coefficients, the signal is reconstructed, and the denoised effective signal can be obtained.

[0051] (ii) Moving average filtering Based on wavelet denoising, a moving average filter is used to further smooth the time-series data. The moving average filter is implemented recursively, and the calculation formula is as follows:

[0052] In the formula, For the first The output value of each data point after being filtered by the moving average; For the first Original grinding process data; For the first Original process data; This is the length of the sliding filter window; This serves as an index for the data sequence. The formula uses a recursive approach to achieve smooth filtering of time-series data, effectively reducing random interference and device noise in the data and improving data stability.

[0053] In this embodiment, the length of the sliding filter window is taken as... .

[0054] (III) Outlier Removal Based on the 3σ Criterion The 3σ criterion is used to detect and remove outliers in the data after wavelet denoising and moving average filtering. The basic principle of the 3σ criterion is that for data that follows a normal distribution, approximately 99.73% of the data points fall within the range of the mean ± 3 standard deviations. Data points outside this range can be considered outliers.

[0055] The specific operation is as follows: calculate the mean of the data sequence. and standard deviation Will satisfy Data points were identified as outliers and removed. The missing data points were then filled using linear interpolation.

[0056] Specific implementation methods for feature extraction and core parameter selection: (I) Multidimensional Feature Extraction The feature extraction subsystem performs multi-dimensional feature extraction in the time domain, frequency domain, and time-frequency domain on the preprocessed core key process parameters.

[0057] Time-domain characteristics include: mean (reflecting the central trend of the signal), variance (reflecting the dispersion of the signal), peak value (reflecting the amplitude limit of the signal), and root mean square (reflecting the effective value of the signal, reflecting the energy level of the signal).

[0058] Frequency domain features are obtained by performing a Fast Fourier Transform (FFT) on the preprocessed signal, including characteristic frequencies (reflecting the dominant frequency components of the signal) and power spectral density (reflecting the distribution of signal energy in the frequency domain).

[0059] The time-frequency domain features employ wavelet entropy. Wavelet entropy characterizes the time-frequency complexity of a signal and is a core feature for determining the grinding state. Based on wavelet transform, wavelet entropy treats the energy of wavelet coefficients as a probability distribution and calculates their information entropy value to quantify the complexity and uncertainty of the signal in the time-frequency domain.

[0060] (II) Core Parameter Screening Based on the floating oil seal processing quality judgment standard, key quality indicators such as surface roughness, flatness, parallelism, and dimensional accuracy of the sealing surface are identified, and correlation analysis between multi-source process parameters and key quality indicators is carried out.

[0061] Pearson correlation coefficient was used to analyze the correlation between each process parameter and key quality indicators. The formula for calculating the Pearson correlation coefficient is:

[0062] In the formula, Pearson correlation coefficient is used to characterize the degree of linear correlation between two sets of data; This represents the total number of data samples; The first Group data samples; Data sequences and The arithmetic mean; This is the sample sequence number. The range of values ​​for this coefficient is... The closer the absolute value is to 1, the stronger the linear correlation between the two sets of data.

[0063] This indicator is used to quantitatively evaluate the quality and feature correlation of preprocessed data, explore the inherent mapping relationship between process parameters and key quality indicators, identify the core key process parameters that play a decisive role in processing quality, eliminate redundant parameters, and reduce the computational complexity of subsequent analysis.

[0064] Combining principal component analysis (PCA) with data dimensionality reduction projects the high-dimensional feature space onto the low-dimensional principal component space, further reducing the feature dimensionality while preserving the main information.

[0065] Based on the characteristics of the grinding process, process features strongly correlated with the processing quality of the sealing surface (such as grinding pressure fluctuation amplitude, temperature change rate, vibration dominant frequency offset, etc.) are extracted to construct a floating oil seal processing quality feature dataset containing equipment status, process parameters, and quality-related features, providing effective feature support for online quality identification.

[0066] Specific implementation of the deep learning-based online processing quality identification model: (I) CNN+LSTM Fusion Network Model Structure Using the constructed quality feature dataset as training samples, and combining the process characteristics of floating oil seal grinding, a deep learning model based on the fusion of convolutional neural network (CNN) and long short-term memory network (LSTM) was designed.

[0067] The feature extraction formula for CNN convolutional layers is:

[0068] In the formula, Indicates the first In the layer network Output characteristics of each neuron; The number of network layers; Indicates the relationship with the first A set of neighboring neurons in which each neuron has connections; For the first From the layer The connection weight coefficients of each neuron; For the first The first in the layer Output characteristics of each neuron; For the first Layer The bias term corresponds to each neuron. This formula is used to realize the transfer and weighted fusion of feature information between network layers, so as to complete the adaptive extraction and expression of data features (where... (This represents the convolution operation).

[0069] The formula for calculating the pooling layer in a CNN is:

[0070] In the formula, For the first Layer The output features after pooling at each feature location; The spatial coordinates of the pooled receptive field; The pooled receptive field area is defined; For the first Layer corresponding spatial location The input feature value is calculated at the receptive field region. This formula achieves local feature filtering and dimensionality reduction by taking the maximum value of the feature value within the receptive field region, retaining key feature information and reducing data redundancy.

[0071] The overall calculation process of the model is as follows:

[0072] In the formula, This refers to the prediction result or feature vector output after computation by the network model. Preprocessed data for the floating oil seal grinding process input into the network model; It is a convolutional neural network operation unit used to extract local spatial features of input data; The feature dimension reconstruction operation is used to convert the convolution output features into a temporal dimension format that is adapted to the input of the recurrent network. It is a long short-term memory network computing unit used to mine temporal correlation features in data sequences; This is a fully connected layer operation used to map and integrate the fused high-dimensional features, ultimately outputting the target prediction result.

[0073] The specific structure of the CNN network is as follows: The input layer receives multidimensional tensor data from the quality feature dataset; the first convolutional layer uses a 3×3 convolutional kernel, has 32 output channels, a stride of 1, and uses ReLU activation; the first pooling layer uses 2×2 max pooling; the second convolutional layer uses a 3×3 convolutional kernel, has 64 output channels, a stride of 1, and uses ReLU activation; the second pooling layer uses 2×2 max pooling. After convolution and pooling operations, the output feature map is converted to a temporal dimension format through a reshape operation.

[0074] The specific structure of the LSTM network is as follows: it contains two layers of LSTM units, each with a hidden state dimension of 128, and uses a gating mechanism (input gate, forget gate, output gate) to mine long-term temporal dependencies in the data sequence.

[0075] The fully connected layer flattens the high-dimensional temporal features output by the LSTM, and then passes them sequentially through a Dropout layer (with a dropout rate of 0.5) and a fully connected layer (with the output dimension being the number of quality state categories). Finally, the probability distribution of each category is output through the Softmax activation function.

[0076] (II) Model Training Supervised training is conducted using a training sample set constructed from historical processing data. The labels of the training samples are marked based on the detection results of offline high-precision inspection equipment (coordinate measuring machine, surface roughness meter, optical flat), and are divided into categories such as quality qualified, surface roughness out of tolerance, flatness not up to standard, parallelism not up to standard, dimensional deviation, and sealing surface defects.

[0077] The cross-entropy loss function is used as the optimization objective for model training. The Adam optimizer is used for gradient descent optimization. The initial learning rate is set to 0.001, the batch size is set to 64, the training epochs are set to 100, and an early stopping strategy is used to prevent overfitting.

[0078] (III) Model Deployment and Online Recognition The trained model is deployed to the edge computing terminal of the quality inspection system to realize local real-time computation of feature extraction and quality judgment, ensuring that the response speed of the inspection system matches the grinding process cycle.

[0079] In the actual processing, the collected real-time data is preprocessed and feature extracted before being input into the deployed model. The model performs real-time calculations locally and outputs the quality status judgment result.

[0080] The formula for calculating the comprehensive evaluation index is as follows:

[0081] In the formula, This is a comprehensive evaluation index value; These are the weighting coefficients for each evaluation indicator; The individual indicators (corresponding to surface roughness, flatness, parallelism, and dimensional accuracy) participating in the comprehensive evaluation are weighted and summed to obtain the final comprehensive evaluation result. The weight coefficients are determined based on the actual working conditions of the floating oil seal, using the analytic hierarchy process (AHP) combined with expert experience.

[0082] (iv) Visualized early warning A visual monitoring interface for the quality inspection system is established to display the equipment's operating status, core process parameters (grinding disc speed, grinding pressure, processing temperature, etc.), and processing quality judgment results in real time. When an abnormal quality state is detected, the system automatically issues an audible and visual warning signal and simultaneously stores the abnormal data and operation logs, providing data support for subsequent quality traceability and algorithm optimization.

[0083] Specific implementation methods for knowledge graph construction: (I) Collection and organization of expert knowledge The system systematically summarizes the experience of enterprise experts, comprehensively collecting expert knowledge on the causes, control methods, and process parameter thresholds of common quality problems in floating oil seal grinding. Common quality problems include, but are not limited to: flatness deviation, surface scratches, insufficient sealing performance, excessive surface roughness, non-compliance with parallelism standards, and dimensional deviations.

[0084] The causes of each type of quality problem include, but are not limited to: abrasive passivation, grinding pressure fluctuations, abnormal processing temperature, unstable grinding disc speed, workpiece positioning misalignment, and inappropriate abrasive particle size.

[0085] The control measures corresponding to each type of cause include, but are not limited to: adjusting the grinding pressure, adjusting the grinding disc speed, changing the abrasive, adjusting the processing time, and calibrating the limit tooling.

[0086] Complete the structured organization and standardized definition of knowledge, and clarify the judgment criteria for each quality problem, the quantitative threshold of each cause, and the parameter range of each control scheme.

[0087] (II) Knowledge Graph Construction Using "quality problems - failure causes - key parameters - control schemes" as the core logic, a knowledge graph for quality control in floating oil seal processing is constructed.

[0088] The knowledge graph uses a triplet structure to store knowledge, specifically in the form of: (quality problem, correlation, cause), (cause, correlation, key parameter), (key parameter, correlation, control plan). For example: (flatness deviation, caused by..., grinding pressure fluctuation), (grinding pressure fluctuation, influencing parameters, grinding pressure value), (grinding pressure value, control measures, adjust the electronic control adjusting nut to the target pressure).

[0089] Knowledge graphs are stored using graph databases (such as Neo4j), enabling the digitization, visualization, and rapid retrieval of expert experience. Graph query languages ​​allow for quick path retrieval from quality issues to control measures.

[0090] (III) Deterministic Factor Reasoning The deterministic factor algorithm is used for logical reasoning. The deterministic factor (CF) is a method for handling uncertain knowledge representation and reasoning, which expresses the uncertain relationship between evidence and conclusions through a confidence measure.

[0091] The formula for calculating deterministic factor inference is:

[0092] In the formula, Assumptions for the conclusion The deterministic factor; As evidence Credibility (range of values) Positive values ​​indicate that the evidence supports the conclusion, negative values ​​indicate that the evidence contradicts the conclusion, and 0 indicates that the evidence is irrelevant to the conclusion. As evidence Supporting conclusions The rule strength (range of values) (This is determined by experts based on experience); the conclusion certainty factor is updated by taking the maximum value of the product of multiple reasoning rules.

[0093] During the reasoning process, the system uses the currently detected quality anomaly type as initial evidence, performs forward or backward reasoning along the association path of the knowledge graph, calculates the deterministic factor of each candidate cause, selects the cause with the highest deterministic factor as the fault diagnosis result, and outputs the corresponding control plan.

[0094] Specific implementation methods of intelligent control algorithms driven by data-knowledge fusion: (I) Establishment of the parameter mapping model Based on tens of thousands of historical processing data and corresponding quality inspection results, a multiple linear regression algorithm is used to establish a quantitative correlation between process parameters and workpiece quality indicators.

[0095] The formula for calculating the multiple linear regression model is as follows:

[0096] In the formula, These are predicted values ​​for quality indicators; This is the regression intercept term; These are the regression coefficients corresponding to each input variable; The independent variables (grinding process parameters, including grinding pressure, grinding disc speed, processing temperature, processing time, etc.) input to the model are used to fit and predict the target variables (surface roughness, flatness, parallelism, dimensional accuracy, etc.) through linear combination.

[0097] The regression coefficients are estimated using the least squares method to minimize the sum of squared residuals between the predicted and actual values.

[0098] (II) Data-Knowledge Fusion Regulation The design incorporates a data- and knowledge-driven intelligent control algorithm that integrates real-time data from the quality inspection system with expert experience from a knowledge graph. This algorithm establishes a quantitative mapping model between process parameters and quality indicators, clarifying the impact patterns and control thresholds of each key parameter on quality.

[0099] When the online quality identification model determines that the quality status deviates from the acceptable range or there is a risk of exceeding tolerances, the control algorithm initiates inference: The first step is to retrieve the corresponding causes and control measures in the knowledge graph based on the type of the current quality problem, and determine the most likely cause of failure through deterministic factor reasoning. The second step is to combine real-time process data with a multiple linear regression quantitative mapping model to calculate the target parameter adjustment required to eliminate quality deviations based on the current process parameters. The third step is to combine the qualitative reasoning results of the knowledge graph with the quantitative calculation results of the regression model, and output the final control decision through a weighted fusion strategy, including the adjusted grinding pressure target value, grinding disc speed target value, and expected adjustment time. The fourth step involves sending the adjusted process parameters to the electric servo actuator (electric servo pressure regulating mechanism and electric servo drive assembly) of the improved grinding device via an electrical interface, thereby achieving adaptive and precise adjustment of the parameters.

[0100] (III) Closed-loop control An active control mechanism was developed. The algorithm receives real-time status feedback from the quality inspection system. When it detects that the quality status deviates from the qualified range or there is a risk of exceeding tolerance, it automatically infers and outputs the optimal process parameter adjustment scheme, which is then sent to the improved grinding device through the electrical interface to achieve adaptive and precise adjustment of parameters.

[0101] Construct a closed-loop control system covering the entire process from equipment processing to data acquisition, quality identification, parameter adjustment, equipment reprocessing, and quality verification. This enables intelligent and unmanned control of floating oil seal grinding, completely eliminating reliance on manual experience.

[0102] During the adjustment process, the system continuously monitors the changing trend of the quality status. If the quality still fails to meet the standard after one adjustment, multiple rounds of iterative adjustment will be carried out until the quality meets the standard or the preset upper limit of the number of adjustments is reached.

[0103] (iv) Verification of the regulatory effect The effectiveness of the control was evaluated using the root mean square error (RMSE). The formula for calculating the root mean square error is:

[0104] In the formula, RMSE Root mean square error is used to measure the accuracy of model predictions. The number of samples; For the first The actual observed values ​​of each sample (measured values ​​from offline high-precision detection equipment); The RMSE (Record Value for Predicted Values) is the model's predicted value for the corresponding sample (the judgment value of the online quality identification system). It is calculated by taking the square root of the mean square of the deviations between the actual and predicted values, reflecting the overall prediction error level of the model. A smaller RMSE value indicates a greater consistency between the online recognition results and the offline detection results, and thus higher model accuracy.

[0105] A quality inspection platform was established, and offline quality inspection of the adjusted floating oil seals was carried out using professional testing equipment such as coordinate measuring machines, surface profilometers, and sealing performance test benches to verify the effectiveness and accuracy of the algorithm control.

[0106] We conduct iterative optimization of the algorithm, continuously refining its reasoning logic and parameter thresholds (including the strength of each rule in the knowledge graph) based on actual processing data and quality inspection results. (Updates, re-estimation of regression model coefficients, and adjustment of the weights of comprehensive evaluation indicators, etc.) improve the generalization ability and control accuracy of the algorithm, and ensure that the algorithm is adapted to the processing needs of floating oil seals of different specifications and working conditions.

[0107] Complete implementation process of online closed-loop control technology: S1 Workpiece clamping and equipment initialization Place the standard floating oil seal metal sealing ring workpiece for engineering machinery at the center of the limiting fixture 8. Secure the workpiece position with the annular limiting iron ring, confirming a firm positioning. Start the equipment; the system automatically completes a hardware self-test (including the communication status of each sensor, the readiness status of the electric servo actuator, and the operating status of the edge computing terminal) and a communication system self-test (including the industrial Ethernet connection status and the communication link status between the data acquisition terminal and the backend server). After passing the self-test, the system enters the processing state.

[0108] S2 Multi-Source Data Synchronous Acquisition The operator clicks the start button, and the equipment begins grinding. Multi-source sensor acquisition components collect six types of process parameters around the clock at a sampling frequency of no less than 10kHz: grinding pressure, equipment vibration, processing temperature, acoustic emission signal, grinding disc speed, and processing energy consumption. The collected data is uploaded in real-time to the multi-source online quality monitoring system via industrial Ethernet, and a time-series database synchronously stores the data in real time.

[0109] S3 Data Processing and Feature Extraction The data preprocessing subsystem processes the collected raw data in the following ways: wavelet denoising (Daubechies wavelet, 5-level decomposition) is used to filter out electromagnetic interference and mechanical noise, followed by moving average filtering (window length). Smooth time-series data and use the 3σ criterion to remove anomalous data.

[0110] The feature extraction subsystem extracts the time-domain features (mean, variance, peak value, root mean square), frequency-domain features (characteristic frequency, power spectral density), and time-frequency-domain features (wavelet entropy) of the preprocessed data. Through Pearson correlation coefficient analysis and principal component analysis, it selects the core parameters that are strongly correlated with the quality of the sealing end face and constructs the quality feature vector of the current workpiece.

[0111] S4 Online Quality Identification The selected core feature parameters are input into a CNN+LSTM fusion deep learning model deployed on an edge computing terminal. The model performs real-time computation locally, completing a quality assessment every 0.5 seconds and outputting a quality status category (quality acceptable, surface roughness out of tolerance, flatness not up to standard, parallelism not up to standard, dimensional deviation, sealing surface defect, etc.) and a comprehensive evaluation index value. The judgment results are displayed in real time on the visual monitoring interface.

[0112] If the quality is deemed acceptable, the system will maintain the current process parameters and continue operating, with a green normal status indicator displayed on the visual interface.

[0113] S5 Intelligent Closed-Loop Control If the judgment result shows a quality deviation (comprehensive evaluation index) If the value falls below a preset threshold, or the model output is any unqualified category, the system will automatically trigger the control process. The first step is for the control system to retrieve matching fault causes in the knowledge graph based on the type of quality problem—by calculating the CF value of each candidate cause using the deterministic factor algorithm, the cause with the highest CF value is selected as the diagnostic result. The second step involves the parameter mapping module calculating the target parameter adjustment amount (including grinding pressure adjustment amount, grinding disc speed adjustment amount, etc.) required to eliminate the deviation based on the current real-time process parameter values ​​and the deviation amount of quality indicators using a multiple linear regression model. The third step is for the adaptive execution module to convert the target adjustment amount into specific control commands (such as the target pressure value of electric cylinder 2 and the target speed value of servo motor). Fourth, control commands are sent to the electric servo pressure regulating mechanism and drive mechanism via industrial Ethernet to achieve real-time adjustment of grinding pressure and grinding disc speed without stopping the equipment.

[0114] The system continuously monitors the changes in quality status after adjustments. If the quality still fails to meet the standard after one adjustment, the next round of iterative adjustment will be completed within 100ms until the quality meets the standard or the preset maximum number of adjustments is reached.

[0115] S6 Finished Product Discharge and Data Recording After the preset processing time is reached, the equipment automatically stops, and a visual interface indicates that processing is complete. The operator then removes the finished workpiece, completing a single processing cycle.

[0116] The system automatically summarizes the complete data of this processing (including full-process sensor data, quality identification results, control records, final quality indicators, etc.) and stores it in the time series database as data support for subsequent model iteration optimization and process parameter standard formulation.

[0117] System calibration and performance verification experiments: (a) Sensor calibration A dedicated quality inspection platform was established, equipped with standard signal sources and standard sensors. Each type of sensor (pressure sensor, vibration sensor, temperature sensor, acoustic emission sensor, and speed sensor) was calibrated individually. The pressure sensor is calibrated at multiple points (0%, 25%, 50%, 75%, 100% of the range) using a standard piston pressure gauge, and the calibration curve is fitted to correct the sensitivity coefficient. The vibration sensor is calibrated using a standard vibration table, covering a frequency range of 10Hz to 10kHz, to calibrate amplitude linearity and frequency response characteristics. The temperature sensor is calibrated using a standard platinum resistance thermometer, and its temperature range covers 0℃~200℃. The acoustic emission sensor uses a broken lead signal for sensitivity calibration; The speed sensor is calibrated using a standard speed source, and the speed range covers 0 to 3000 rpm.

[0118] (ii) Calibration of electric servo actuator Accuracy calibration of the electric servo pressure regulating mechanism and electric servo drive assembly: By providing a series of target pressure and speed values ​​via electronic control signals, the actual output values ​​are measured using standard pressure sensors and standard tachometers. The input-output relationship curve is then fitted, and the control parameters are corrected to ensure that the execution accuracy meets the design requirements (pressure control accuracy better than ±1%FS, speed control accuracy better than ±0.5%FS).

[0119] (III) Validation of the online quality identification model The hold-out method was used to validate the model's performance: the labeled historical dataset was divided into training and test sets in a 7:3 ratio. After training on the training set, the model was evaluated on the test set. Evaluation metrics included accuracy, precision, recall, and F1 score.

[0120] The online recognition results of the model are compared and analyzed with the detection results of offline high-precision detection equipment (coordinate measuring machine, surface roughness meter, optical flat), and the recognition accuracy is calculated. When the online recognition accuracy is lower than 95%, the model training phase is returned to adjust the network hyperparameters (learning rate, batch size, number of network layers, etc.) or expand the training sample set until the accuracy reaches the required level.

[0121] (iv) Performance verification of the entire system Batch processing experiments were conducted on floating oil seals of different specifications (different diameters, different materials) and different grinding process parameters to verify the stability and reliability of the system under all working conditions.

[0122] Comparative experimental scheme: Select floating oil seal workpieces of the same model, and process 500 pieces each using traditional grinding equipment and the device of this embodiment, and compare the processing quality and production efficiency.

[0123] The results of the comparative experiment are as follows:

[0124] Experiments have shown that this invention significantly improves processing accuracy, product qualification rate, and production efficiency. The eccentricity of the floating oil seal sealing end face is stably controlled from 0.22–0.45 mm in traditional equipment to 0.10–0.18 mm, meeting the long-term reliable operation requirement of eccentricity not exceeding 0.2 mm; the product qualification rate is increased from 82% in traditional equipment to over 98%.

[0125] Adaptation and machining of floating oil seals of different specifications: For floating oil seal workpieces of different diameters (such as 100mm, 150mm, 200mm, etc.), replace the corresponding annular limit fixture 8 to ensure the positioning accuracy of the workpiece during the grinding process. Pre-enter the standard process parameters (recommended grinding pressure, recommended grinding disc speed, recommended processing time, etc.) and quality judgment thresholds for each workpiece specification into the system database.

[0126] Before processing, the operator selects the corresponding workpiece specification on the visual interface, and the system automatically loads the standard process parameters and judgment thresholds for that specification. During processing, the system performs online identification and closed-loop control based on the quality requirements of that specification.

[0127] For floating oil seal workpieces made of different materials (such as high-chromium cast iron and chromium-molybdenum alloy cast iron), the system has preset corresponding abrasive selection schemes and process parameter compensation coefficients. The system automatically adjusts the benchmark values ​​and control thresholds of parameters such as grinding pressure and rotation speed according to the material type to ensure that workpieces of different materials can meet the specified surface quality requirements.

[0128] After each batch of processing is completed, the system automatically summarizes the data for that batch and updates the process parameter database for that specification / material workpiece. When the accumulated data reaches a preset threshold, the system automatically triggers model retraining, using the newly added data to optimize the regression coefficients of the multiple regression model and the regularity of each rule in the knowledge graph. This allows for continuous iterative optimization of process parameter standards, gradually bringing the processing quality closer to the optimal state.

[0129] Remote monitoring and data traceability: The system interfaces with the enterprise's Manufacturing Execution System (MES) via industrial Ethernet to enable remote real-time monitoring of processing data. Authorized personnel can access a visual monitoring interface via a remote terminal to view the real-time operating status of each device, current processing parameters, quality judgment results, and equipment warning information.

[0130] All raw sensor data, preprocessed feature data, quality identification results, and control records from the entire processing are stored in a time-series database for a period of no less than three years. When quality problems occur, technicians can quickly retrieve the complete data chain of the entire processing process for a workpiece using timestamps and workpiece batch numbers, enabling them to trace the root cause from the final quality defect to the processing abnormality.

[0131] The system automatically generates quality reports for each batch of processing, including pass rate statistics, quality index distribution charts, and process parameter trend charts, providing data support for the quality management system and meeting the traceability requirements of quality management systems such as ISO9001 for the production process.

[0132] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A floating oil seal intelligent grinding system, characterized in that, include: The grinding device body includes a frame (1), a grinding disc (6), at least one limiting fixture (8), a grinding execution mechanism, and a driving mechanism; the grinding disc (6) is mounted on the frame (1) and is used to place the workpiece; the limiting fixture (8) is located inside the grinding disc (6); the grinding execution mechanism is arranged above the limiting fixture (8); the driving mechanism is mounted on the bottom of the frame (1) and is used to drive the grinding disc (6) to rotate; a multi-source sensor acquisition component is integrated on the grinding device body and is used to collect multi-dimensional process parameters in real time during the grinding process; The online quality inspection system is communicatively connected to the multi-source sensor acquisition component and is used to identify the workpiece processing quality online based on the multi-dimensional process parameters. The intelligent control system is connected to the online quality detection system, the drive mechanism, and the multi-source sensor acquisition components, respectively, and is used to adaptively adjust the grinding process parameters based on the quality identification results.

2. The floating oil seal intelligent grinding system according to claim 1, characterized in that, The multi-source sensing acquisition component includes a pressure sensor, a vibration sensor, a temperature sensor, an acoustic emission sensor, a speed sensor, and an energy consumption acquisition module, used to simultaneously acquire multi-dimensional process parameters such as grinding pressure, equipment vibration, processing temperature, acoustic emission signal, grinding disc speed, and processing energy consumption; the multi-source sensing acquisition component is laid with shielded cables and establishes a communication connection with the online quality detection system through an industrial Ethernet.

3. The intelligent grinding system for floating oil seals according to claim 1, characterized in that, The online quality detection system includes a data preprocessing subsystem, a feature extraction subsystem, and a quality identification subsystem. The data preprocessing subsystem uses wavelet transform for signal denoising, combines moving average filtering to smooth time-series data, and employs the 3σ criterion to remove outliers. The feature extraction subsystem extracts time-domain, frequency-domain, and time-frequency-domain features from the data, and uses correlation analysis to select core feature parameters related to processing quality, constructing a quality feature dataset. The quality identification subsystem uses a CNN+LSTM fusion network model, with the quality feature dataset as training samples. The CNN+LSTM fusion network model sequentially passes through a convolutional neural network, feature reconstruction, a long short-term memory network, and a fully connected unit to complete local spatial feature extraction, temporal dimension transformation, temporal feature mining, and quality result output, respectively.

4. The intelligent grinding system for floating oil seals according to claim 3, characterized in that, The quality identification subsystem is deployed on an edge computing terminal to realize local real-time computation of feature extraction and quality determination.

5. The intelligent grinding system for floating oil seals according to claim 1, characterized in that, The intelligent control system includes a knowledge graph module, a parameter mapping module, and an adaptive execution module. The knowledge graph module uses a triplet structure of "quality problem - fault cause - key parameter - control scheme" to store and process knowledge, and uses a deterministic factor algorithm to complete fault logic reasoning. The parameter mapping module uses a multiple linear regression algorithm to establish a quantitative correlation between process parameters and workpiece quality indicators; The adaptive execution module sends control commands to the drive mechanism based on the quality identification results and the calculation results of the parameter mapping module, and automatically adjusts the grinding process parameters.

6. The intelligent grinding system for floating oil seals according to claim 1, characterized in that, The grinding device body also includes a pressure regulating mechanism, which is located above the grinding execution mechanism and is used to regulate the grinding pressure applied to the workpiece by the grinding execution mechanism; the pressure regulating mechanism includes an electric cylinder (2) and an electrically controlled adjusting nut (3), and the driving mechanism is a servo motor.

7. A floating oil seal intelligent grinding closed-loop control process using the system described in any one of claims 1 to 6, characterized in that, Includes the following steps: S1: Clamp the floating oil seal workpiece in the limiting fixture (8) and start the whole system; S2: Multi-source sensor acquisition components collect multi-dimensional process parameters in real time during the grinding process; S3: The online quality inspection system preprocesses and extracts features from the collected data to identify the current processing quality status of the workpiece; S4: If the processing quality deviates, the intelligent control system matches the corresponding control strategy through the knowledge graph and calculates the target adjustment amount of the process parameters through the parameter mapping model. S5: The intelligent control system sends control commands to the drive mechanism to adjust the grinding process parameters in real time, and the equipment does not need to be stopped during the parameter adjustment process. S6: Repeat steps S2 to S5 to form a closed loop of the entire process of "data acquisition - quality identification - parameter adjustment - reprocessing" until the grinding process is completed.

8. The closed-loop control process according to claim 7, characterized in that, In step S3, the preprocessing includes: using wavelet transform to denoise the signal, using moving average filtering to smooth the data, and using the 3σ criterion to remove abnormal data.

9. The closed-loop control process according to claim 7, characterized in that, In step S3, the feature extraction includes: extracting time-domain, frequency-domain, and time-frequency-domain features, and using the Pearson correlation coefficient analysis method to screen core feature parameters related to processing quality.

10. The closed-loop control process according to claim 7, characterized in that, The quality deviations mentioned in step S4 include at least one of the following: excessive surface roughness of the sealing end face, non-compliance with flatness standards, non-compliance with parallelism standards, dimensional deviations, abnormal vibration caused by abrasive passivation, and abnormal processing temperature; the closed-loop response speed of the intelligent control system from quality identification to parameter adjustment is controlled within 100ms.

Images