Fault early warning method for brush wheel shaft group of PCB wet process automatic voltage regulation polish-brush machine

Abstract

The embodiment of the invention provides a fault early warning method for a brush wheel shaft group of a PCB wet process automatic voltage regulation polish-brush machine, and the method comprises the steps: S1, collecting fault signals of the brush wheel shaft group, the fault signals comprising a vibration signal, a drive motor three-phase current signal, a bearing temperature signal, and a polish-brush pressure signal; s2, performing feature extraction on the fault signal to obtain a high-dimensional feature vector; s3, inputting the high-dimensional feature vector into a pre-trained deep learning diagnosis model for feature-level fusion and state recognition, wherein the diagnosis model outputs a comprehensive score representing the health state of the brush wheel shaft group; and S4, according to whether the comprehensive score exceeds a preset dynamic early warning threshold, whether an early warning signal is sent out is judged, and the dynamic early warning threshold can be automatically adjusted according to the real-time operation condition of the equipment. According to the method, the detection sensitivity of weak fault features is improved, the early warning accuracy under different working conditions is ensured, the phenomena of false alarm and missing alarm are reduced, and the service life of equipment is prolonged.

Description

Technical Field

[0001] This application relates to the field of brush wheel shaft assembly maintenance for brush grinding machines, and in particular to a fault early warning method for brush wheel shaft assembly of an automatic voltage-adjusting brush grinding machine in PCB wet process. Background Technology

[0002] The wet process in PCB manufacturing is a critical step, and the automatic pressure-adjusting brush machine, as the core equipment for surface treatment, directly affects product quality and production efficiency through the stable operation of its brush wheel shaft assembly. In actual production environments, the brush wheel shaft assembly operates under high-speed, heavy-load, and wet corrosive conditions for extended periods, making its mechanical components prone to progressive damage. If potential faults are not detected in time, they can lead not only to batch product defects but also to cascading equipment damage, resulting in significant economic losses.

[0003] While existing technologies have attempted to incorporate vibration or temperature sensors for condition monitoring, these solutions are mostly limited to simple alarms based on single parameter thresholds. They lack in-depth integration and intelligent analysis of multi-dimensional operational information, resulting in insufficient accuracy in early warnings and a coexistence of false alarms and missed alarms. This is particularly true for automatic pressure-regulating brush mills, where shaft assembly failures are often gradual failures caused by the interaction of mechanical, electrical, and process parameters. Existing methods struggle to effectively predict their remaining service life and cannot provide forward-looking decision support for planned maintenance, posing a challenge to achieving truly predictive maintenance.

[0004] Therefore, there is an urgent need in this field for an innovative method that can deeply integrate multi-source information and has early and accurate diagnostic and predictive capabilities to improve equipment reliability and ensure production quality and efficiency. Summary of the Invention

[0005] This application provides a method for early warning of brush wheel shaft assembly faults in an automatic voltage-adjusting brush grinding machine for wet PCB manufacturing. This method, through the synergistic effect of multi-source signal fusion and intelligent diagnostic model, can achieve accurate early warning of brush wheel shaft assembly faults, effectively overcoming the limitations of traditional single-signal monitoring, significantly improving the detection sensitivity of weak fault features, and ensuring the accuracy of early warning under different operating conditions through an adaptive threshold mechanism. This greatly reduces false alarms and missed alarms, extends equipment lifespan, and reduces unplanned downtime, thereby effectively reducing overall operation and maintenance costs while improving product quality consistency.

[0006] Firstly, a method for early warning of brush wheel shaft assembly failure in an automatic voltage-adjusting brushing machine for wet PCB manufacturing is provided, the method comprising:

[0007] S1: Collect fault signals of the brush wheel shaft assembly, including vibration signals, three-phase current signals of the drive motor, bearing temperature signals, and brush pressure signals;

[0008] S2: Perform feature extraction on the fault signal to obtain a high-dimensional feature vector;

[0009] S3: Input the high-dimensional feature vector into a pre-trained deep learning diagnostic model for feature-level fusion and state recognition. The diagnostic model outputs a comprehensive score characterizing the health status of the brush wheel shaft group.

[0010] S4: Determine whether to issue a warning signal based on whether the comprehensive score exceeds the preset dynamic warning threshold. The dynamic warning threshold can be automatically adjusted according to the real-time operating conditions of the equipment.

[0011] It should be understood that vibration signals, three-phase current signals from the drive motor, bearing temperature signals, and brush pressure signals constitute a complementary system capable of comprehensively sensing the brush wheel shaft assembly from four dimensions: mechanical dynamics, electrical load, thermodynamic state, and process execution effect. Vibration signals can sensitively capture early mechanical shocks to components such as bearings and gears; current signals reflect torque fluctuations and abnormal loads throughout the entire transmission chain from the motor to the brush wheel; temperature signals directly reveal the accumulation of frictional heat due to lubrication failure or excessive wear, serving as a key indicator of fault severity; and pressure signals are directly related to process quality, effectively distinguishing between equipment failure and parameter misalignment. This multimodal fusion strategy overcomes the limitations of single-parameter monitoring, which is susceptible to interference and has a high false alarm rate under complex wet process conditions. Through cross-validation between signals, it achieves comprehensive and accurate diagnosis of potential faults, from early warning and cause identification to severity assessment.

[0012] In conjunction with the first aspect, in some implementations of the first aspect, step S2 includes:

[0013] S201: Perform envelope demodulation analysis on the vibration signal to extract the frequency band energy related to the bearing fault characteristic frequency;

[0014] S202: Perform a fast Fourier transform on the current signal to analyze the amplitude changes of its specific harmonic components; S203: Calculate the temperature rise gradient of the temperature signal;

[0015] S204: Calculate the dynamic fluctuation variance of the pressure signal.

[0016] In conjunction with the first aspect, in some implementations of the first aspect, step S201 includes:

[0017] The optimal frequency band containing the bearing fault characteristic frequencies was determined using resonant frequency band screening technology.

[0018] The signal within the optimal frequency band is envelope demodulated using Hilbert transform to obtain the envelope spectrum; the frequency band energy value corresponding to the bearing fault characteristic frequency in the envelope spectrum is calculated.

[0019] In conjunction with the first aspect, in some implementations of the first aspect, the diagnostic model includes a one-dimensional dilated convolution and an improved covariance mutual attention mechanism; the dilation factor of the one-dimensional dilated convolution is configured to be dynamically adjusted according to the input high-dimensional feature vector; the improved covariance mutual attention mechanism is configured to calculate the covariance matrix between different modal features, obtain the minimized covariance of different modal features, and calculate the weighting coefficients of different features based on the minimized covariance.

[0020] It should be understood that by employing a one-dimensional dilated convolutional layer with dynamically adjusted dilation factors, the receptive field of the model is effectively expanded, enhancing its ability to capture multi-scale fault features. At the same time, combined with an improved covariance mutual attention mechanism, feature decoupling and optimized fusion are achieved by minimizing the covariance between features of different modalities. This significantly improves the model's fusion efficiency and state recognition accuracy for multi-source heterogeneous fault features, thereby enabling more sensitive and reliable detection of early and weak faults.

[0021] In conjunction with the first aspect, in some implementations of the first aspect, the method for dynamically adjusting the dynamic early warning threshold includes:

[0022] Establish a threshold adjustment model with the brush machine operating speed and the current conductivity of the grinding slurry as input variables; adjust the dynamic early warning threshold based on the operating speed and the conductivity of the slurry.

[0023] In conjunction with the first aspect, in some implementations of the first aspect, step S3 includes:

[0024] S301: Based on the features extracted by the diagnostic model, perform fault mode matching and output the most likely fault type and location information;

[0025] S302: Based on the identified fault mode, generate the comprehensive score and a diagnostic report including suggested inspection points and recommended maintenance measures.

[0026] In conjunction with the first aspect, in certain implementations of the first aspect, after the diagnostic model is put into use, an online transfer learning strategy is employed to continuously optimize the diagnostic model, including:

[0027] When the production line changes to a new type of PCB board, causing a change in the operating conditions of the brush wheel shaft assembly, the diagnostic model is fine-tuned using a small amount of sample data under the new operating conditions.

[0028] Adjust some parameters in the diagnostic model to enable it to quickly adapt to new operating conditions and maintain high-precision diagnostic capabilities.

[0029] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes predictive maintenance support:

[0030] Perform time series trend analysis on the comprehensive score to obtain the analysis results;

[0031] Based on the results, the remaining service life of key components of the brush wheel shaft assembly is predicted, and time window suggestions are provided. It should be understood that by introducing predictive maintenance support functionality, combining real-time monitoring with long-term trend analysis, the remaining service life of key components can be accurately predicted based on time-series data with comprehensive scoring, and precise maintenance time window suggestions can be provided. This achieves a shift from passive maintenance to proactive maintenance, effectively extending equipment lifespan, significantly reducing unplanned downtime, and optimizing maintenance resource allocation, thereby reducing overall maintenance costs while ensuring production continuity. Attached Figure Description

[0032] Figure 1 A flowchart illustrating a fault warning method for the brush wheel shaft assembly of an automatic pressure-regulating brush machine in a wet PCB process, provided in this application embodiment.

[0033] Figure 2 The flowchart illustrates a method for extracting high-dimensional feature vectors from fault signals, as provided in this embodiment of the application.

[0034] Figure 3 The flowchart illustrates a comprehensive scoring method for obtaining a characterization of the health status of a brush wheel shaft assembly, as provided in this application embodiment. Detailed Implementation

[0035] The terminology used in the following embodiments is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” and “this” are intended to also include expressions such as “one or more,” unless the context clearly indicates otherwise. It should also be understood that in the following embodiments of this application, “at least one” and “one or more” refer to one, two, or more than two. The term “and / or” is used to describe the relationship between related objects, indicating that three relationships can exist; for example, A and / or B can indicate: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character “ / ” generally indicates that the preceding and following related objects are in an “or” relationship.

[0036] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0037] In PCB wet manufacturing processes, the brush wheel shaft assembly of an automatic pressure-regulating brushing machine operates under high-speed, heavy-load, and wet corrosion conditions for extended periods. This makes its mechanical components susceptible to progressive damage, directly impacting product quality and production efficiency. Existing monitoring solutions largely rely on single-parameter threshold alarms, lacking deep integration of multi-dimensional operational information. This results in insufficient early warning accuracy and frequent false alarms and missed alarms. Particularly concerning is the progressive failure under the combined effects of multiple factors; current technologies struggle to effectively predict remaining lifespan and cannot provide forward-looking decision-making for planned maintenance, posing a significant challenge to predictive maintenance.

[0038] This application provides a method for early warning of brush wheel shaft assembly failure in an automatic pressure-adjusting brush machine for wet PCB manufacturing, which can effectively overcome the above-mentioned problems.

[0039] The technical solutions provided in the embodiments of this application will be described below with reference to the accompanying drawings.

[0040] Figure 1 A flowchart illustrating a fault warning method for the brush wheel shaft assembly of an automatic pressure-regulating brush machine in a wet PCB process, provided in this application embodiment.

[0041] refer to Figure 1 In some examples, the method includes:

[0042] S1: Collect fault signals of the brush wheel shaft assembly, including vibration signals, three-phase current signals of the drive motor, bearing temperature signals, and brush pressure signals;

[0043] S2: Perform feature extraction on the fault signal to obtain a high-dimensional feature vector;

[0044] S3: Input the high-dimensional feature vector into a pre-trained deep learning diagnostic model for feature-level fusion and state recognition. The diagnostic model outputs a comprehensive score characterizing the health status of the brush wheel shaft group.

[0045] S4: Determine whether to issue a warning signal based on whether the comprehensive score exceeds the preset dynamic warning threshold. The dynamic warning threshold can be automatically adjusted according to the real-time operating conditions of the equipment.

[0046] Figure 2 The flowchart illustrates a method for extracting high-dimensional feature vectors from fault signals, as provided in this embodiment of the application.

[0047] refer to Figure 2 In some examples, step S2 includes:

[0048] S201: Perform envelope demodulation analysis on the vibration signal to extract the frequency band energy related to the bearing fault characteristic frequency;

[0049] S202: Perform a fast Fourier transform on the current signal to analyze the amplitude changes of its specific harmonic components; S203: Calculate the temperature rise gradient of the temperature signal;

[0050] S204: Calculate the dynamic fluctuation variance of the pressure signal.

[0051] In some examples, step S201 includes:

[0052] The optimal frequency band containing the bearing fault characteristic frequencies was determined using resonant frequency band screening technology.

[0053] The signal within the optimal frequency band is envelope demodulated using Hilbert transform to obtain the envelope spectrum; the frequency band energy value corresponding to the bearing fault characteristic frequency in the envelope spectrum is calculated.

[0054] In some examples, the diagnostic model includes a one-dimensional dilated convolution and an improved covariance mutual attention mechanism; the dilation factor of the one-dimensional dilated convolution is configured to be dynamically adjusted based on the input high-dimensional feature vector; the improved covariance mutual attention mechanism is configured to calculate the covariance matrix between different modal features, obtain the minimum covariance of different modal features, and calculate the weighting coefficients of different features based on the minimum covariance. In one possible implementation, in the one-dimensional dilated convolution module, the input high-dimensional feature vector is first analyzed through a lightweight feedforward network, and an optimal dilation factor sequence is dynamically generated based on its feature distribution. This sequence is determined by a differentiable search algorithm to ensure that the receptive field can adapt to fault features at different scales. In the covariance mutual attention module, the model first calculates the covariance matrix of multimodal feature vectors such as vibration, current, and temperature. It then minimizes the covariance between different modal features by introducing an orthogonal constraint loss function, achieving feature decoupling. Next, based on the decoupled features, mutual attention scores are calculated. A learnable transformation matrix maps the covariance-minimized features to a unified semantic space. Finally, a softmax function is used to generate weighted coefficients for each feature, completing effective feature fusion. The two modules work collaboratively through residual connections, preserving multi-scale spatiotemporal features while ensuring deep fusion of multi-source information.

[0055] In some examples, the dynamic adjustment method for the dynamic warning threshold includes:

[0056] Establish a threshold adjustment model with the brush machine operating speed and the current conductivity of the grinding slurry as input variables; adjust the dynamic early warning threshold based on the operating speed and the conductivity of the slurry.

[0057] Figure 3 The flowchart illustrates a comprehensive scoring method for obtaining a characterization of the health status of a brush wheel shaft assembly, as provided in this application embodiment.

[0058] refer to Figure 3 In some examples, step S3 includes:

[0059] S301: Based on the features extracted by the diagnostic model, perform fault mode matching and output the most likely fault type and location information;

[0060] S302: Based on the identified fault mode, generate the comprehensive score and a diagnostic report including suggested inspection points and recommended maintenance measures.

[0061] In one possible implementation, the feature vectors extracted by the diagnostic model are first compared with the pre-stored fault feature templates in the database using cosine similarity calculation. The K-nearest neighbor algorithm is then used to determine the most matching fault type, and fault location is achieved based on the distribution differences of features at different positions in the shaft group. Subsequently, the diagnostic report generation module automatically generates a diagnostic report containing illustrations of specific inspection locations, suggested maintenance measures, and priorities, based on the identified fault type and its severity, combined with rules from the equipment maintenance knowledge base. The comprehensive score is generated by inputting factors such as feature matching degree, fault severity index, and equipment runtime into a trained scoring prediction network. This network adopts a multilayer perceptron structure, which can comprehensively consider various factors to output a health score and preset corresponding maintenance suggestion templates for different score ranges.

[0062] In some examples, after the diagnostic model is put into use, an online transfer learning strategy is used to continuously optimize the diagnostic model, including:

[0063] When the production line changes to a new type of PCB board, causing a change in the operating conditions of the brush wheel shaft assembly, the diagnostic model is fine-tuned using a small amount of sample data under the new operating conditions.

[0064] Adjust some parameters in the diagnostic model to enable it to quickly adapt to new operating conditions and maintain high-precision diagnostic capabilities.

[0065] In some examples, the method also includes predictive maintenance support:

[0066] Perform time series trend analysis on the comprehensive score to obtain the analysis results;

[0067] Based on the results, the remaining service life of key components of the brush wheel shaft assembly is predicted, and time window suggestions are provided. In one possible implementation, this predictive maintenance method first synchronously collects vibration, current, temperature, and pressure signals of the brush wheel shaft assembly, and then extracts multi-dimensional indicators such as the envelope spectrum characteristics of the vibration signals and the harmonic characteristics of the current to construct a comprehensive health status score; subsequently, time series analysis is performed on the score based on algorithms such as Gaussian mixture models to distinguish between normal, sub-healthy, and faulty states of the equipment; finally, predictive algorithms such as particle filtering are used to extrapolate the performance degradation trajectory, estimate the remaining service life, and generate maintenance time window suggestions.

[0068] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. Any equivalent modifications or variations made by those skilled in the art based on the content disclosed in the present invention should be included within the scope of protection set forth in the claims.

Claims

1. A method for early warning of brush wheel shaft assembly faults in an automatic voltage-adjusting brushing machine for wet PCB manufacturing, characterized in that, The method includes: S1: Collect fault signals of the brush wheel shaft assembly, including vibration signals, three-phase current signals of the drive motor, bearing temperature signals, and brush pressure signals; S2: Perform feature extraction on the fault signal to obtain a high-dimensional feature vector; S3: Input the high-dimensional feature vector into a pre-trained deep learning diagnostic model for feature-level fusion and state recognition. The diagnostic model outputs a comprehensive score characterizing the health status of the brush wheel shaft group. S4: Determine whether to issue a warning signal based on whether the comprehensive score exceeds the preset dynamic warning threshold. The dynamic warning threshold can be automatically adjusted according to the real-time operating conditions of the equipment.

2. The method according to claim 1, characterized in that, Step S2 includes: S201: Perform envelope demodulation analysis on the vibration signal to extract the frequency band energy related to the bearing fault characteristic frequency; S202: Perform a fast Fourier transform on the current signal to analyze the amplitude changes of its specific harmonic components; S203: Calculate the temperature rise gradient of the temperature signal; S204: Calculate the dynamic fluctuation variance of the pressure signal.

3. The method according to claim 2, characterized in that, Step S201 includes: The optimal frequency band containing the bearing fault characteristic frequencies was determined using resonant frequency band screening technology. The envelope spectrum is obtained by using the Hilbert transform to demodulate the signal within the optimal frequency band. Calculate the frequency band energy value in the envelope spectrum corresponding to the bearing fault characteristic frequency.

4. The method according to claim 2, characterized in that, The diagnostic model includes a one-dimensional dilated convolution and an improved covariance mutual attention mechanism; the dilation factor of the one-dimensional dilated convolution is configured to be dynamically adjusted according to the input high-dimensional feature vector; The improved covariance mutual attention mechanism is configured to calculate the covariance matrix between different modal features, obtain the minimum covariance of different modal features, and calculate the weighting coefficients of different features based on the minimum covariance.

5. The method according to claim 4, characterized in that, The dynamic adjustment method for the dynamic early warning threshold includes: Establish a threshold adjustment model with the brush machine operating speed and the current conductivity of the grinding slurry as input variables; The dynamic early warning threshold is adjusted based on the operating speed and the slurry conductivity.

6. The method according to claim 5, characterized in that, Step S3 includes: S301: Based on the features extracted by the diagnostic model, perform fault mode matching and output the most likely fault type and location information; S302: Based on the identified fault mode, generate the comprehensive score and a diagnostic report including suggested inspection points and recommended maintenance measures.

7. The method according to claim 6, characterized in that, After the diagnostic model is put into use, an online transfer learning strategy is used to continuously optimize the diagnostic model, including: When the production line changes to a new type of PCB board, causing a change in the operating conditions of the brush wheel shaft assembly, the diagnostic model is fine-tuned using a small amount of sample data under the new operating conditions. Adjust some parameters in the diagnostic model to enable it to quickly adapt to new operating conditions and maintain high-precision diagnostic capabilities.

8. The method according to claim 7, characterized in that, The method also includes predictive maintenance support: Perform time series trend analysis on the comprehensive score to obtain the analysis results; Based on the results, the remaining service life of the key components of the brush wheel shaft assembly is predicted, and a time window suggestion is provided.

Images