Online detection system with wear assessment function for cement ball mill

Abstract

The present invention discloses an online oil detection system and method for cement ball mills, relating to the technical field of industrial equipment status monitoring and fault diagnosis. The system includes a data acquisition module, which uses a multi-channel vibration sensor array to synchronously acquire data on the positions of a ball mill bearing seat, a gearbox, and a drive shaft. An original vibration signal matrix containing multi-component coupling information is obtained through high-frequency sampling. If the sampling frequency is lower than ten times the wear characteristic frequency, the sampling parameters are automatically adjusted until the requirements of the Nyquist theorem are met. A fault warning and model adaptation module adapts to changes in equipment operating conditions by updating model parameters in real time. The online oil detection system and method for cement ball mills effectively improve equipment operating reliability and maintenance efficiency, extend service life, reduce maintenance costs, and provide data support for preventive maintenance decisions for ball mills.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial equipment status monitoring and fault diagnosis, and in particular to an online detection system with a wear assessment function for a cement ball mill. Background Art

[0002] Ball mills are core equipment in heavy industries such as mineral processing, cement production, and the chemical industry. Their stable operation is directly related to the efficiency and safety of the entire production line. Monitoring the wear of key components such as bearings and gears is crucial for preventing equipment failures and reducing maintenance costs. With the increasing degree of industrial automation, the need for real-time monitoring of equipment health is becoming increasingly urgent. Traditional wear monitoring methods rely primarily on periodic downtime inspections and manual judgment. This passive maintenance model not only causes production interruptions and losses but also fails to accurately assess the actual wear process of components. While existing online monitoring technologies can collect basic signals such as vibration and temperature, their signal analysis capabilities are limited under complex operating conditions, often failing to provide accurate wear assessments and reliable fault warnings. The complex vibration signals generated by the simultaneous operation of multiple key components within a ball mill are coupled to each other, creating an extremely complex signal environment. When multiple wear modes coexist, such as bearing rolling element wear, gear tooth spalling, and drive shaft imbalance, the characteristic signals of each component experience severe aliasing in the frequency and time domains. This signal aliasing directly complicates wear feature extraction, making it difficult for traditional analysis methods to accurately isolate wear information of specific components from the complex signal. Crucially, the signal characteristics of different wear stages exhibit nonlinear variations. Early, weak wear signals are often masked by the noise of normal equipment operation. By the time the wear signal becomes noticeable enough, the component may be nearing failure. This nonlinear nature of the wear evolution process makes it extremely difficult to develop an accurate wear state assessment model. Existing methods are unable to quantitatively assess the extent of wear and reliably predict its development trends. Summary of the Invention

[0003] The purpose of the present invention is to provide an online detection system for cement ball mill with a wear assessment function to solve the problems existing in the prior art.

[0004] To achieve the above object, the present invention provides the following technical solution: an online detection system for cement ball mill with wear assessment function, the system comprising:

[0005] The data acquisition module uses a multi-channel vibration sensor array to synchronously collect data on the positions of the ball mill's bearing seat, gearbox, and drive shaft. High-frequency sampling is used to obtain a raw vibration signal matrix containing multi-component coupling information. If the sampling frequency is less than ten times the wear characteristic frequency, the sampling parameters are automatically adjusted until the Nyquist theorem requirements are met.

[0006] The wear assessment and trend analysis module obtains real-time wear assessment results for condition monitoring based on the original vibration signal matrix. It uses sliding time window technology to perform trend analysis on the wear assessment results within a continuous time period, fits the wear evolution curve using the least squares method, calculates the wear development rate and acceleration based on the curve slope and second-order derivative, and obtains wear evolution law parameters for predictive analysis.

[0007] The fault warning and model adaptation module establishes a fault warning mechanism based on the parameters of the wear evolution law. If the wear development rate exceeds the safety threshold or the predicted remaining life is lower than the maintenance cycle, a warning signal is triggered and maintenance recommendations are generated. The model parameters are updated in real time to adapt to changes in equipment operating conditions.

[0008] Preferably, the wear assessment and trend analysis module obtains real-time wear assessment results for condition monitoring based on the original vibration signal matrix, including:

[0009] Preprocessing operations are performed on the original vibration signal matrix. A bandpass filter is used to remove power frequency interference and high-frequency noise. The random noise component is eliminated through the time domain synchronous averaging algorithm to obtain a purified multi-channel vibration signal data set for subsequent analysis and processing.

[0010] Preferably, the wear assessment and trend analysis module obtains real-time wear assessment results for condition monitoring based on the original vibration signal matrix and further includes:

[0011] The purified multi-channel vibration signal dataset is subjected to blind source separation using the independent component analysis algorithm. The coupled composite signal is decomposed into mutually independent source signal components according to the principle of statistical independence. If the number of separated signal components exceeds a preset threshold, the kurtosis criterion is used to screen out the main components related to wear.

[0012] Preferably, the wear assessment and trend analysis module obtains real-time wear assessment results for condition monitoring based on the original vibration signal matrix and further includes:

[0013] The time-frequency domain features are extracted for the independent source signal components, and the wavelet packet decomposition algorithm is used to obtain the energy distribution characteristics of different frequency bands. By calculating the energy entropy value and relative energy ratio of each frequency band, a multidimensional feature vector set reflecting the wear status of different components is obtained.

[0014] Preferably, the wear assessment and trend analysis module obtains real-time wear assessment results for condition monitoring based on the original vibration signal matrix and further includes:

[0015] A support vector regression model is established based on a set of multidimensional feature vectors. The radial basis kernel function is used to map the feature space. The regression parameters are trained using historical wear data. If the training error exceeds the preset accuracy requirement, the kernel function parameters and regularization coefficient are adjusted until the model converges.

[0016] Preferably, the wear assessment and trend analysis module obtains real-time wear assessment results for condition monitoring based on the original vibration signal matrix and further includes:

[0017] The support vector regression model is used to quantify the wear degree of the current eigenvector. The wear level of the bearing and gear is determined based on the regression output value. The health status of the component is judged by comparing it with the preset wear threshold, and real-time wear assessment results are obtained for condition monitoring.

[0018] Preferably, the multi-channel vibration sensor array includes at least three channels, which are respectively arranged at the positions of the bearing seat, gear box and transmission shaft of the ball mill.

[0019] Preferably, the fault warning and model adaptation module has a warning level classification mechanism, sets multi-level thresholds according to the wear rate and remaining life prediction results, and outputs warning instructions and maintenance suggestions of different levels.

[0020] Preferably, it also includes a data communication module for transmitting wear assessment results, trend data and warning information to a host monitoring platform or a remote expert system in real time via industrial Ethernet or wireless communication.

[0021] It can be seen from the above technical solution that the present invention has the following beneficial effects:

[0022] This online wear assessment system for cement ball mills uses a multi-channel vibration sensor array to collect raw vibration signals from the bearing seat, gearbox, and drive shaft. Preprocessing and independent component analysis are used to separate the signals. Time-frequency domain features are extracted to construct a support vector regression model for quantitatively assessing the degree of bearing and gear wear. A sliding time window is used to analyze wear evolution trends, establish an early warning mechanism, and generate maintenance recommendations. This system enables real-time monitoring, trend analysis, and fault warning of the wear status of key ball mill components, effectively improving equipment operational reliability and maintenance efficiency, extending service life, and reducing maintenance costs, providing data support for preventive maintenance decisions for ball mills. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a connection diagram of the system modules of the present invention;

[0024] Figure 2 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0026] like Figure 1 As shown, the present invention provides a technical solution: an online detection system with a wear assessment function for a cement ball mill, the system comprising:

[0027] The data acquisition module uses a multi-channel vibration sensor array to synchronously collect data on the positions of the ball mill's bearing seat, gearbox, and drive shaft. High-frequency sampling is used to obtain a raw vibration signal matrix containing multi-component coupling information. If the sampling frequency is less than ten times the wear characteristic frequency, the sampling parameters are automatically adjusted until the Nyquist theorem requirements are met.

[0028] The wear assessment and trend analysis module obtains real-time wear assessment results for condition monitoring based on the original vibration signal matrix. It uses sliding time window technology to perform trend analysis on the wear assessment results within a continuous time period, fits the wear evolution curve using the least squares method, calculates the wear development rate and acceleration based on the curve slope and second-order derivative, and obtains wear evolution law parameters for predictive analysis.

[0029] The fault warning and model adaptation module establishes a fault warning mechanism based on the parameters of the wear evolution law. If the wear development rate exceeds the safety threshold or the predicted remaining life is lower than the maintenance cycle, a warning signal is triggered and maintenance recommendations are generated. The model parameters are updated in real time to adapt to changes in equipment operating conditions.

[0030] The system integrates a multi-channel vibration sensor array to enable simultaneous status monitoring of multiple key components of the ball mill. Each vibration sensor is attached to the bearing housing, gearbox, and drive shaft, capturing minute mechanical vibration changes through high-frequency sampling. During acquisition, the system automatically monitors whether the sampling frequency meets the requirement of at least ten times the wear characteristic frequency. If not, a dynamic sampling adjustment mechanism is triggered to ensure that data sampling meets the Nyquist sampling theorem, thereby avoiding spectral aliasing and information loss.

[0031] The raw vibration signal is organized into a high-dimensional matrix and input into the Wear Assessment and Trend Analysis module. Time-frequency domain analysis techniques such as Fast Fourier Transform (FT) or Wavelet Transform (WT) are first used to extract typical frequency components and identify potential abnormal frequency signatures. Subsequently, the system uses methods based on Empirical Mode Decomposition (EMD) or Ensemble Empirical Mode Decomposition (EEMD) to extract intrinsic mode functions (IMFs) and further analyze wear patterns within the non-stationary signal. Using sliding time window technology, the system dynamically extracts wear state parameters for each time period during continuous operation. By constructing a time series and applying least squares regression fitting to these parameters, the system generates a wear evolution curve.

[0032] By calculating the first-order derivative and the second-order derivative of the curve respectively, the system estimates the current wear rate and acceleration in real time, extracts the trend change parameters of wear development, and provides reliable input for predictive analysis. Next, the fault warning and model adaptation module performs a threshold comparison analysis on the parameters of the wear evolution law. When a sharp increase in the wear rate is detected or the predicted remaining life is lower than the set safety boundary, the warning mechanism is automatically triggered and corresponding maintenance suggestions are given. In order to cope with the continuous changes in the operating status of the equipment, the module also integrates a model adaptation function, using a self-updating parameter mechanism or an online learning strategy, such as recursive least squares or incremental training algorithms, to ensure that the model always reflects the latest operating characteristics of the equipment, ensuring the accuracy and robustness of the system prediction and evaluation.

[0033] This system uses a technical approach based on multi-source high-frequency vibration signal analysis and is suitable for monitoring ball mills with complex coupling structures. The data acquisition module is equipped with a three-channel accelerometer array, which is installed in key locations such as the ball mill's bearing seat, gearbox, and drive shaft to simultaneously and synchronously capture the original vibration signals generated during the equipment's operation. The system sets the lower limit of the initial sampling frequency to more than ten times a reference frequency, which corresponds to the highest wear characteristic frequency that the equipment may produce. If the sampling frequency is insufficient to meet this condition during actual operation, the system will automatically increase the sampling rate to ensure that the collected signal fully reflects the actual operating conditions and avoid spectrum aliasing or loss of key information.

[0034] During the signal processing phase, the system first performs segmented analysis on the collected vibration signal, splitting the entire time-domain signal into multiple time segments and extracting the frequency characteristics of each segment to observe how the vibration intensity and frequency change over time. To further extract nonlinear or non-stationary characteristics, the system uses a technique called ensemble empirical mode decomposition (EMD) to decompose the complex signal into multiple independent frequency modes and extract the key information contained in each frequency band.

[0035] The system then sets a sliding time window, which can range from ten seconds to a complete ball mill cycle, and calculates a characteristic value within each window, which measures the wear of the equipment. Over time, the system accumulates the characteristic values ​​from multiple time windows and arranges them in chronological order, forming a data series that can be used for trend fitting.

[0036] The system performs trend modeling on this data series, using a mathematical fitting method to approximate it as a quadratic curve that changes over time. Each time point in this curve corresponds to a predicted wear value; the slope at that point indicates the rate of wear growth at that moment, known as the wear rate; and the curvature of the curve at that point indicates the trend of increasing wear, known as the wear acceleration.

[0037] The resulting wear rate and acceleration serve as key parameters for evaluating equipment health and are then fed into the next stage. The system sets a safety limit to define the upper limit of the equipment's wear rate under normal operating conditions. If the actual wear rate exceeds this safety limit, or if the remaining service life of the equipment, as estimated based on wear trends, falls short of the preset maintenance interval, the system proactively issues an alert and generates a maintenance recommendation.

[0038] To ensure the system remains effective despite changes in equipment load, fluctuations in operating status, and even equipment aging, an adaptive model update module is included. This module automatically updates the parameters of the wear trend model based on new data collected in real time, aligning the system's analysis capabilities with the equipment's current operating status, thereby ensuring the continued accuracy and real-time nature of the early warning mechanism.

[0039] This invention achieves real-time status monitoring and precise wear prediction of key ball mill components through the synchronous acquisition and analysis of high-frequency vibration coupling signals. A multi-channel array design enhances the signal's spatial information representation capabilities, and EEMD analysis improves the modeling accuracy of nonlinear and nonstationary signals. The system incorporates adaptive sampling and modeling mechanisms, significantly improving the timeliness of fault identification and prediction accuracy, effectively reducing the incidence of sudden failures, extending equipment life, and enhancing the scientific nature of maintenance decisions. Compared to traditional single-point detection, this system offers significant improvements in performance, stability, and intelligence.

[0040] The wear assessment and trend analysis module obtains real-time wear assessment results for condition monitoring based on the original vibration signal matrix. This includes preprocessing operations based on the original vibration signal matrix, using a bandpass filter to remove power frequency interference and high-frequency noise, and eliminating random noise components through a time-domain synchronous averaging algorithm. The resulting purified multi-channel vibration signal data set is used for subsequent analysis and processing.

[0041] First, the system synchronously collects raw vibration signals from multiple channels of the ball mill's multi-channel vibration sensor array. These signals contain information about the actual state of the equipment during operation, but are also contaminated by environmental interference and noise. To improve the accuracy of subsequent analysis, the system first performs bandpass filtering on these raw signals. The lower limit frequency of the bandpass filter is generally set at 10 Hz to remove 50 Hz power frequency interference from the power grid and low-frequency background signals caused by low-frequency mechanical vibrations. The upper limit frequency is generally set within 5 kHz to remove electromagnetic interference and high-frequency electronic noise. The goal of this filtering process is to retain valid mid-frequency signals that are highly correlated with wear characteristics.

[0042] The filtered signal may still contain some non-periodic interference caused by operational disturbances. Therefore, the system further incorporates synchronous averaging technology. This involves aligning multiple consecutive vibration signals according to the position of each sensor and then averaging the data at the same time point. This method effectively eliminates random components, as non-periodic noise approaches zero over multiple averagings. The periodic signal components, on the other hand, are superimposed and amplified, providing a clearer representation of the equipment's stable operating state.

[0043] To perform this synchronous averaging, the system first measures the ball mill's spindle speed and calculates the time required for each rotation. It then extracts data segments representing multiple complete cycles from the original signal, overlays them, and calculates the average. Typically, the system selects five to ten cycles for averaging, though the specific number can be adjusted based on signal stability and processing accuracy requirements.

[0044] After the aforementioned filtering and synchronous averaging, the system generates a purified set of multi-channel vibration signals. These signals significantly reduce interference from environmental and system noise, resulting in greater stability and interpretability. This purified set of signals is then fed into the subsequent state extraction module, where key characteristic parameters for determining wear severity are further extracted through methods such as envelope demodulation, Hilbert transform, and wavelet decomposition. These parameters serve as input for equipment condition monitoring and wear trend prediction analysis.

[0045] By introducing two signal preprocessing technologies, bandpass filtering and time-domain synchronous averaging, the purification ability and feature retention of vibration signals are significantly improved. The bandpass filter can effectively filter out power frequency interference and high-frequency noise, ensuring that only frequency components closely related to mechanical wear are retained in the signal, thereby improving the accuracy and stability of subsequent analysis results. The synchronous averaging technology enhances the extraction effect of periodic wear signals during equipment operation by averaging multiple cycles, while suppressing the random noise introduced by irregular disturbances and occasional interference, significantly improving the signal-to-noise ratio in the feature extraction process. Compared with traditional single time-domain or frequency-domain filtering processing methods, the multi-step signal purification process provided by the present invention has stronger robustness and is suitable for cement ball mill status monitoring scenarios under various complex operating conditions. As a result, the wear assessment process is more sensitive and accurate, which can effectively avoid false alarms and missed alarms, lay a reliable data foundation for subsequent trend analysis and fault warning, improve the system's predictive ability and operational stability, and extend the equipment's trouble-free operation time.

[0046] The wear assessment and trend analysis module obtains real-time wear assessment results for condition monitoring based on the original vibration signal matrix. It also uses the independent component analysis algorithm to perform blind source separation on the purified multi-channel vibration signal data set. According to the principle of statistical independence, the coupled composite signal is decomposed into mutually independent source signal components. If the number of separated signal components exceeds a preset threshold, the kurtosis criterion is used to screen out the main components related to wear.

[0047] After signal preprocessing, the independent component analysis algorithm is introduced to perform blind source separation operation to identify the intrinsic components closely related to mechanical wear from the multi-channel coupled signals.

[0048] At this stage, the cleaned multi-channel vibration signal dataset can be represented as an observation signal matrix, consisting of time series recorded by multiple vibration sensors, referred to as a mixed signal. According to blind source separation theory, this mixed signal can be viewed as a linear combination of several source signals undergoing an unknown mixing process. The basic model is expressed as follows: the mixed signal is equal to the product of an unknown mixing matrix and several independent source signals.

[0049] The core goal of independent component analysis (ICA) is to recover these source signals, that is, to obtain a set of separation matrices that maximize the statistical independence of the recovered signals. Common independence measures include strategies such as minimizing information entropy, maximizing negentropy, and enhancing non-Gaussianity. In this system, the FastICA algorithm is used as the implementation path, and its optimization goal is to maximize the negentropy function to extract the source components.

[0050] The approximate form of negative entropy is that the negative entropy value is equal to the statistical difference between the observed signal and the Gaussian signal under a certain nonlinear function mapping. Its specific form is the expected value difference constructed using nonlinear functions (such as hyperbolic tangent, exponential function, etc.), which is used to evaluate independence.

[0051] During the FastICA iteration process, by maximizing the negative entropy approximation function, the separation vector corresponding to each independent component is gradually estimated, and finally the source signal sequence is obtained.

[0052] After obtaining all independent components, the system introduces the kurtosis criterion as a screening metric to identify components significantly associated with mechanical wear. Kurtosis is a statistic that describes the "sharpness" of a signal's probability distribution and quantifies the degree to which a signal deviates from a Gaussian distribution. It is defined as follows: kurtosis is equal to the signal's fourth-order central moment divided by the square of the second-order central moment, minus a constant of three.

[0053] The higher the kurtosis value, the sharper the signal peak and the stronger the impact, which is common in mechanical impact wear signals.

[0054] When the kurtosis value is close to three, it means that the signal is close to Gaussian distribution and the information contribution is low.

[0055] The system presets a kurtosis threshold, such as ten or fifteen. If the kurtosis value of an independent component is higher than the threshold, it indicates that it may contain sudden, strongly nonlinear wear information, and the component is retained as the main analysis object; the remaining components are excluded to avoid redundant data interfering with the analysis.

[0056] Finally, the filtered high-kurtosis source signals are sent to the wear trend modeling module to complete subsequent curve fitting, rate calculation, warning generation and other processes.

[0057] By introducing the independent component analysis algorithm, the system can effectively identify potential independent sources in the vibration signal, that is, the intrinsic responses generated by multiple coupled components. Compared with traditional methods based on separate channel analysis or frequency filtering, blind source separation technology has the ability to automatically recover wear-related components from complex mixed signals. It does not rely on pre-known signal models or physical assumptions, which greatly improves the sensitivity and breadth of fault diagnosis. In particular, combined with the kurtosis screening mechanism, the present invention can effectively eliminate background stable signals and retain only abnormal vibration modes with impact or nonlinear enhancement characteristics, thereby focusing on key signals that are truly related to the wear state. This processing method improves the accuracy of wear trend modeling, reduces the misjudgment rate and the risk of missed reports, and enables the system to cope with actual industrial scenarios with complex equipment structures and strong coupling interference, with stronger adaptability and technical practicality.

[0058] The wear assessment and trend analysis module obtains real-time wear assessment results for condition monitoring based on the original vibration signal matrix. It also includes time-frequency domain feature extraction for independent source signal components, and uses wavelet packet decomposition algorithm to obtain energy distribution characteristics of different frequency bands. By calculating the energy entropy value and relative energy ratio of each frequency band, a multidimensional feature vector set reflecting the wear status of different components is obtained.

[0059] After purifying the raw vibration signal and separating and filtering its independent components, the system further performs feature extraction on the filtered independent source signal components to comprehensively reflect different wear types and component conditions. This process uses a wavelet packet decomposition algorithm to perform multi-scale time-frequency analysis of the signal, extracting frequency band energy characteristics and entropy information to construct a multidimensional state feature vector.

[0060] The specific steps are as follows: Wavelet packet decomposition processing, for each independent source signal, use wavelet packet decomposition to divide it into multi-level frequency bands. Unlike traditional wavelet decomposition that only further decomposes low-frequency signals, wavelet packet technology can completely decompose high-frequency and low-frequency parts at the same time, thereby achieving uniform coverage of the entire frequency range. Set the decomposition level of wavelet packet decomposition to L, and divide the original signal into 2 L Sub-bands, each representing the energy content within a specific frequency range. The number of decomposition layers is typically determined based on the system sampling rate and the expected fault frequency band. For example, if the sampling rate is 10,000 Hz, choosing three decomposition layers yields eight frequency bands, each approximately 1,250 Hz wide.

[0061] Band energy calculation, the energy of each sub-band signal is based on the result of its two-norm calculation, specifically the sum of the squares of the amplitudes of all sample points. Let the signal corresponding to the j-th band be d j , its energy E j The calculation method is: square the vibration amplitude of all sampling points in the frequency band and then sum them to get the energy value. The sum of the energy of all frequency bands is E total , by calculating the proportion of each frequency band energy to the total energy P j =E j / E total , and obtain the relative energy distribution, which is used to judge the concentration of vibration energy and the distribution of fault frequency bands.

[0062] Energy entropy calculation is introduced to further measure the degree of frequency distribution chaos and spectral spread. Energy entropy is calculated by multiplying the energy ratio of each frequency band by its logarithm, then summing the results and negating the sign to obtain the total energy entropy. A larger value indicates a more dispersed energy distribution, indicating a higher complexity of the vibration signal. A lower entropy value indicates that the energy is concentrated in a few frequency bands, often corresponding to localized structural wear or impact failure.

[0063] A multidimensional feature vector is constructed, combining the relative energy value of each frequency band with the total energy entropy. This vector's dimension is equal to the number of frequency bands plus one, representing a complete wear feature set including the energy ratios and entropies of all frequency bands. This constitutes the input sample used for model training or state recognition.

[0064] The number of wavelet decomposition layers controls the number of frequency bands and analysis accuracy, and is usually set to three to four layers, depending on the sampling rate and spectrum range;

[0065] The sub-band energy is the total vibration energy in each frequency band, which is calculated by summing the squares of the amplitudes;

[0066] The relative energy ratio is the proportion of each frequency band energy to the total energy, which is obtained by dividing the sub-band energy by the total energy;

[0067] Energy entropy represents the entropy value of the energy distribution of the frequency band, and uses Shannon entropy calculation logic to evaluate complexity and concentration.

[0068] By introducing wavelet packet decomposition and frequency band energy feature extraction methods, the system can perform refined analysis of vibration responses within different frequency ranges, significantly enhancing the ability to identify multi-component and multi-type wear behaviors. In particular, the introduction of the energy entropy index enables the system to not only determine the energy concentration interval, but also identify changes in signal complexity, effectively distinguishing local faults from overall vibration anomalies. Compared with traditional single-band analysis methods, the multi-dimensional feature vectors provided by the solution of the present invention cover a wider range of frequency information and have stronger distinguishing and expression capabilities, and are suitable for supporting machine learning model training or threshold-driven judgment. Its processing process does not rely on manual rule setting and is highly automated. It is suitable for long-term, unattended equipment status monitoring scenarios in actual production environments, significantly improving diagnostic accuracy and system practicality.

[0069] The wear assessment and trend analysis module obtains real-time wear assessment results for condition monitoring based on the original vibration signal matrix. It also establishes a support vector regression model based on a multidimensional feature vector set, uses a radial basis kernel function to map the feature space, and trains the regression parameters using historical wear data. If the training error exceeds the preset accuracy requirement, the kernel function parameters and regularization coefficient are adjusted until the model converges.

[0070] Building on the multidimensional wear feature vectors extracted through wavelet packet decomposition, this system further introduces a nonlinear regression model based on machine learning to predict future wear trends. This model uses support vector regression. Its core concept is to build a functional model that can be used for future predictions by learning the correspondence between wear features in historical samples and the actual wear state.

[0071] Specifically, the support vector regression model maps the input multidimensional feature vector into a higher-dimensional feature space and then searches for an optimal hyperplane within that space, ensuring that the predicted output is as close to the true value as possible while maintaining an acceptable error tolerance. To achieve this, the system uses a kernel function called a radial basis function to map the original input into a high-dimensional space. This kernel function constructs a similarity metric based on the distance relationship between input features, thereby enhancing the model's ability to model nonlinear relationships.

[0072] During model training, the system first prepares a set of historical data, including multidimensional feature vectors extracted from different time periods and the corresponding actual wear levels as labels. The system uses this data to train a regression model, constructing a function in a high-dimensional space that estimates wear values ​​with minimal error. During training, the system sets an error tolerance to ensure a certain degree of deviation in the model's predictions. A penalty factor is also introduced to balance model complexity and training error.

[0073] After initial model training, if evaluation results indicate that the prediction error exceeds the preset accuracy requirement, for example, if the average error exceeds five percentage points, the system will initiate an automatic parameter adjustment mechanism to adjust the kernel width parameter and the error penalty coefficient. The kernel width affects the model's sensitivity to input changes. A larger kernel width results in a smoother prediction curve, suitable for situations where trends change slowly; a smaller kernel width improves the model's responsiveness to sudden changes. The penalty coefficient adjusts the tolerance for error; a larger value results in a greater tendency for the model to reduce the error at each sample point.

[0074] The system gradually tries different parameter combinations, uses methods such as cross-validation to compare the prediction accuracy under each combination, and finally selects the parameter combination that minimizes the prediction error as the final configuration of the model.

[0075] The trained model can be used to predict wear trends corresponding to the equipment's current state in real time, enabling early detection of abnormal wear and optimizing maintenance strategies. By continuously inputting new feature vectors, the model continuously outputs updated predictions, supporting dynamic decision-making within the system.

[0076] The present invention establishes a nonlinear function framework with high-precision prediction capabilities by introducing a support vector regression model in the wear assessment and trend analysis module. Through historical data training, the model can accurately capture the complex relationship between the characteristic vector and the actual wear state, and realize forward-looking prediction of future wear trends. Compared with the traditional linear regression method, SVR combined with the radial basis kernel function has stronger nonlinear modeling capabilities, and can still maintain stable prediction accuracy in scenarios with large fluctuations in equipment operating status and frequent load changes. By automatically adjusting the kernel function parameters and regularization coefficients, the system can achieve rapid convergence of training errors while avoiding overfitting, and enhance the adaptability to different working conditions and data scales. Ultimately, the integration of this model significantly improves the reliability of equipment operating status prediction, provides a solid data foundation for preventive maintenance and intelligent decision-making, and improves the overall operating efficiency and safety of the system.

[0077] The wear assessment and trend analysis module obtains real-time wear assessment results for condition monitoring based on the original vibration signal matrix. It also uses a support vector regression model to quantify the degree of wear of the current eigenvector, determines the wear level of bearings and gears based on the regression output value, and judges the health status of the components by comparing with the preset wear threshold to obtain real-time wear assessment results for condition monitoring.

[0078] After the wear trend modeling phase is complete, the system uses a support vector regression model to perform real-time wear prediction on the collected feature vectors, quantifying the wear severity of key equipment components. The specific process is as follows: The system first processes vibration data from a specific time period through preprocessing, wavelet decomposition, and feature extraction, generating a multi-dimensional feature vector. This feature vector is then fed into the trained support vector regression model for calculation. The model, having learned the mapping between input features and wear severity during training, outputs a continuous value as the current predicted wear level. This predicted value represents the wear severity of a specific equipment component (such as a bearing or gear). The system then compares this predicted value against a set of predefined wear level ranges. Typically, these levels are categorized into several ranges: for example, a predicted value between 0 and 0.3 indicates mild wear, 0.3 to 0.6 indicates moderate wear, 0.6 to 0.9 indicates heavy wear, and values ​​exceeding 0.9 indicate severe wear. These wear level thresholds are set based on extensive practical operation and maintenance experience and historical data analysis, and have clear physical meaning. For example, in a bearing scenario, slight wear may correspond to surface microcracks or poor lubrication, moderate wear may be the early stages of raceway spalling, and severe wear usually indicates imminent failure. Once the system completes the wear level judgment, it will automatically determine the health status of the equipment based on the range to which it belongs. If it is judged to be slight wear, the system only records the current status; if it is moderate or severe, maintenance recommendations are generated; if it is a serious level, an alarm is immediately triggered, and the control system can be linked to implement shutdown protection or send a remote notification. The entire process does not rely on human intervention, and all judgment criteria are derived from data-driven models and threshold comparison logic. It is highly automated and standardized, ensuring that the system can stably and accurately provide real-time wear assessment results under various operating conditions to serve equipment status monitoring and proactive maintenance management.

[0079] By applying the support vector regression model to the continuous quantitative calculation of the degree of wear, the present invention not only realizes the real-time evaluation of components such as bearings and gears under complex operating conditions, but also establishes a standardized wear grade classification system, so that the system has a clear and executable basis for intelligent judgment. Compared with the traditional extensive monitoring mode based on empirical thresholds, the present invention uses a data-driven approach to continuously model the wear trend and convert it into a grade classification output, which has higher accuracy, sensitivity and engineering interpretability. It can detect small changes at an early stage, accurately identify the direction of trend development, and trigger corresponding strategies according to preset logic, such as alarms, recording or linkage maintenance, greatly improving the automation and intelligence level of equipment monitoring. At the same time, the grade output method is concise and clear, adapted to the industrial site's operation and maintenance management needs for the "red, yellow, and green" classification status, and improves the system's operability and promotion value in actual engineering scenarios.

[0080] The multi-channel vibration sensor array includes at least three channels, which are respectively arranged at the positions of the bearing seat, the gear box and the transmission shaft of the ball mill.

[0081] To comprehensively monitor the vibration status of key parts of the ball mill, this system incorporates a vibration sensor array with at least three channels in the data acquisition module. These sensors are located at the mill's bearing housing, gearbox, and drive shaft, corresponding to typical mechanical vibration sources. The goal is to synchronously capture coupled vibration signals from multiple components.

[0082] The first channel's sensor, installed near the bearing housing, primarily detects low- and medium-frequency vibrations caused by rolling element or raceway fatigue, insufficient lubrication, or hardened grease. The second channel, located on the gearbox housing, collects high-frequency or impact vibrations caused by periodic shocks, meshing clearance changes, and gear breakage that may occur during gear meshing. The third channel, installed near the drive shaft, senses dynamic changes caused by torque fluctuations, axis misalignment, or coupling loosening, reflecting the system's torsional and eccentric characteristics.

[0083] All three channels are configured for high-frequency sampling and controlled by a unified trigger signal, ensuring that each signal remains synchronized in time. This means that each signal is sampled at the exact same time, preventing coupling errors caused by data asynchrony.

[0084] The sampling frequency is determined by the target analysis frequency and, in principle, must be significantly higher than the equipment's most significant wear characteristic frequency. In practical applications, the system typically sets the sampling rate between 5,000 and 25,000 samples per second to capture high-frequency shocks or localized faults. For example, when analyzing a signal with a maximum vibration frequency of 1,000 Hz, the sampling frequency should be set to at least 10,000 Hz to ensure sufficient information recovery.

[0085] The collected vibration signals from the three channels are stored in chronological order and form a structured dataset. Each signal corresponds to an independent time series, and the three sets of data are aligned on the time axis. This data structure directly serves as the input for subsequent processing modules such as filtering, feature extraction, blind source separation, and trend modeling.

[0086] To match the mechanical rigidity and expected vibration amplitude of different installation locations, the system configures each channel's gain differently. For example, the drive shaft, where vibration amplitude is relatively low, is suitable for a higher gain; whereas the gearbox, where strong impact signals may occur, requires a medium gain to avoid signal saturation.

[0087] In summary, the rational layout and parameter configuration of the multi-channel array ensure that the system can obtain stable, comprehensive, and high-precision raw vibration signals under different operating conditions, providing solid data support for subsequent intelligent diagnosis and prediction.

[0088] By deploying a multi-channel vibration sensor array at the three core locations of the bearing seat, gearbox, and drive shaft, the system can obtain comprehensive dynamic data representing the operating status of different components. This structure is significantly superior to the single-point monitoring method and can effectively capture coupled signals from different sources of the equipment, thereby supporting subsequent blind source separation, wavelet energy analysis, and trend modeling. The multi-channel design improves the spatial resolution of the data, helps analyze the correlation and timing characteristics between vibration signals, and provides a solid foundation for fault source location. Especially in structures where bearings and gearboxes are jointly driven, this design can identify the composite vibration modes generated by mutual influence, significantly improving diagnostic accuracy and response capabilities. In addition, the array deployment solution has structural flexibility and compatibility, and is suitable for ball mill systems of different specifications and different working conditions. It has good on-site adaptability and promotion value.

[0089] The fault warning and model adaptation module has a warning level classification mechanism, sets multi-level thresholds based on the wear rate and remaining life prediction results, and outputs different levels of warning instructions and maintenance suggestions.

[0090] In the fault warning and model adaptation module, to enhance the system's sensitivity to equipment operating conditions and tailor its response, the present invention implements a multi-level warning classification mechanism based on wear rate and remaining life predictions. This mechanism automatically outputs different levels of warning instructions and corresponding maintenance recommendations based on a dual assessment of the equipment's current state evolution rate and remaining available time.

[0091] First, the system continuously monitors the vibration characteristics of the equipment during operation and calculates the current wear rate using a trend analysis model. The wear rate refers to the rate of change in the degree of wear per unit time. It is calculated by comparing the change in the equipment's wear index in adjacent time periods and dividing it by the time interval to obtain the average rate of change over that period. The system also uses the current trend curve to extrapolate into the future, estimating the time it will take for the equipment to reach the set end-of-wear threshold, assuming the current trend continues. This is the remaining lifespan.

[0092] The system makes a level judgment based on the two key indicators mentioned above, namely wear rate and remaining life. To achieve refined management, the system divides warnings into three levels: level one is warning level, level two is serious level, and level three is critical level. The judgment logic is as follows:

[0093] When the equipment wear rate is low and the remaining life is predicted to be long, the system determines that the current equipment status is good and only issues a low-level reminder or logs;

[0094] When the wear rate increases or the remaining life has entered the medium- to short-term range, the system determines that the current state has certain risks, issues a medium-level warning, and recommends arranging maintenance;

[0095] If the equipment wear rate has significantly accelerated, or the remaining life has fallen below the minimum tolerance time threshold, the system will immediately identify it as a high-level alarm and activate the linkage response mechanism, outputting a complete set of emergency measures including shutdown instructions, emergency maintenance recommendations, and personnel notifications.

[0096] The aforementioned warning level determination process is based on multiple pre-defined numerical thresholds, which are determined based on statistical analysis of historical operating data. For example, the wear rate thresholds might be 1% and 3% per hour, respectively, while the remaining life time limits might be set at 48 hours and 24 hours. During operation, the system compares the current calculated value with these thresholds and automatically matches the corresponding warning level.

[0097] Ultimately, based on the assessment results, the system will output a warning instruction containing a level indicator, textual suggestions, and color labels (e.g., green, orange, and red), allowing operators to clearly understand the current status and take appropriate measures. All warning results are also recorded in a log for subsequent retrospective analysis and model optimization.

[0098] The present invention introduces a multi-level early warning mechanism based on the joint judgment of wear rate and remaining life, so that the system can not only identify abnormal trends in the operating status of the equipment, but also realize dynamic graded response according to the fault evolution speed and the expected remaining life. Compared with the traditional single threshold trigger method, this mechanism has higher response sensitivity and fault discrimination accuracy. By hierarchically managing the wear rate and life prediction results, the system can effectively distinguish between mild fluctuations and serious failures, avoiding false alarms or missed alarms. At the same time, the multi-level response strategy can support the deployment of step-by-step control measures, realize the hierarchical execution of alarms, suggestions and emergency measures, and contribute to the scientific scheduling of production plans and the optimal allocation of operation and maintenance resources. This mechanism improves the foresight and initiative of equipment fault handling, and provides powerful intelligent decision-making support and safety protection for the operation of ball mills under complex working conditions.

[0099] It also includes a data communication module for transmitting wear assessment results, trend data and warning information to a host monitoring platform or remote expert system in real time via industrial Ethernet or wireless communication.

[0100] The online detection system with wear assessment function for cement ball mills of the present invention is provided with a complete data encapsulation, transmission and remote integration mechanism in terms of data communication module, ensuring that the wear assessment results, trend data and warning information can be stably and in real time transmitted to the upper monitoring platform or remote expert system, realizing data closed loop and intelligent collaboration. Specifically, the system first performs structured encapsulation on the data from each module, including the equipment number, acquisition time, current wear assessment value, remaining life prediction value, warning level, wear trend parameters and operating status flag, etc., and adopts common industrial communication formats such as Modbus TCP, OPC UA or MQTT for standardized processing to ensure data compatibility and resolvability. In terms of communication mode, the system can select industrial Ethernet for high-speed and stable transmission according to actual on-site needs, which is suitable for areas with mature wiring; it can also adopt wireless communication methods such as Wi-Fi, 4G or LoRa to adapt to application environments with long distances, difficult wiring or strong equipment mobility. The system defaults to a data push frequency of once every thirty seconds, and also supports event triggering mode based on state changes, with flexible configuration capabilities. To ensure real-time and reliable communication, the system controls the total communication delay to less than two seconds and has built-in cyclic redundancy check and retransmission mechanisms to ensure data integrity and accuracy. In addition, the system supports establishing a stable connection via TCP / IP and allows the remote receiving end to be configured as a visual monitoring platform or AI expert system, thereby providing diagnostic advice in the early stages of a fault and automatically triggering maintenance processes. The integration of this module enables the efficient flow of ball mill operating status information from edge perception to remote collaboration, significantly improving the system's intelligence level, response speed, and management efficiency, and meeting the practical needs of large-scale industrial equipment for reliable remote monitoring and intelligent maintenance.

[0101] By integrating the data communication module, the present invention can realize cross-layer transmission and remote sharing of system-level information, which not only enhances the real-time and responsiveness of the local monitoring system, but also enables the equipment operation status to be transparently presented on the remote platform, providing a real-time data foundation for operation and maintenance scheduling, expert consultation and production management. Compared with the traditional "local data collection, manual reading" mode, the communication mechanism of this system can achieve high-frequency, low-latency, structured data push, and cooperate with standard protocols to achieve seamless docking with mainstream industrial software systems. It is particularly suitable for distributed cement production lines, significantly improving the collaborative perception capability of equipment status, and supporting the provision of intelligent judgment and maintenance suggestions through cloud-based algorithm models in the early stages of fault development to avoid the spread of faults. This function significantly improves the overall intelligence level, response speed and management efficiency of the system, and is particularly suitable for continuous production scenarios with high requirements for equipment reliability and short maintenance windows.

[0102] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An online detection system with wear assessment function for cement ball mill, characterized by: The system comprises: The data acquisition module uses a multi-channel vibration sensor array to synchronously collect data on the positions of the ball mill's bearing seat, gearbox, and drive shaft. High-frequency sampling is used to obtain a raw vibration signal matrix containing multi-component coupling information. If the sampling frequency is less than ten times the wear characteristic frequency, the sampling parameters are automatically adjusted until the Nyquist theorem requirements are met. The wear assessment and trend analysis module obtains real-time wear assessment results for condition monitoring based on the original vibration signal matrix. It uses sliding time window technology to perform trend analysis on the wear assessment results within a continuous time period, fits the wear evolution curve using the least squares method, calculates the wear development rate and acceleration based on the curve slope and second-order derivative, and obtains wear evolution law parameters for predictive analysis. The fault warning and model adaptation module establishes a fault warning mechanism based on the wear evolution law parameters. If the wear development rate exceeds the safety threshold or the predicted remaining life is lower than the maintenance period, a warning signal is triggered and maintenance recommendations are generated. The model parameters are updated in real time to adapt to changes in equipment operating conditions. The wear assessment and trend analysis module also includes: The independent component analysis algorithm is used to perform blind source separation on the multi-channel vibration signal dataset. The coupled composite signal is decomposed into independent source signal components based on the principle of statistical independence. If the number of separated signal components exceeds a preset threshold, the kurtosis criterion is used to screen out the main components related to wear. The time-frequency domain features of the independent source signal components are extracted, and the wavelet packet decomposition algorithm is used to obtain the energy distribution characteristics of different frequency bands. By calculating the energy entropy value and relative energy ratio of each frequency band, a multidimensional feature vector set reflecting the wear status of different components is obtained. A support vector regression model is established based on a multidimensional feature vector set. The radial basis kernel function is used to map the feature space. The regression parameters are trained using historical wear data. If the training error exceeds the preset accuracy requirement, the kernel function parameters and regularization coefficient are adjusted until the model converges. The support vector regression model is used to quantify the wear degree of the current eigenvector. The wear level of the bearing and gear is determined based on the regression output value. The health status of the component is judged by comparing it with the preset wear threshold, and real-time wear assessment results are obtained for condition monitoring.

2. The online detection system with wear assessment function for cement ball mill according to claim 1, characterized in that: The wear assessment and trend analysis module obtains real-time wear assessment results for condition monitoring based on the original vibration signal matrix, including: Preprocessing operations are performed on the original vibration signal matrix. A bandpass filter is used to remove power frequency interference and high-frequency noise. The random noise component is eliminated through the time domain synchronous averaging algorithm to obtain a purified multi-channel vibration signal data set for subsequent analysis and processing.

3. The online detection system with wear assessment function for cement ball mill according to claim 1, characterized in that: The multi-channel vibration sensor array includes at least three channels, which are respectively arranged at the positions of the bearing seat, gear box and transmission shaft of the ball mill.

4. The online detection system with wear assessment function for cement ball mill according to claim 1, characterized in that: The fault warning and model adaptation module has a warning level classification mechanism, sets multi-level thresholds based on wear rate and remaining life prediction results, and outputs warning instructions and maintenance suggestions of different levels.

5. The online detection system with wear assessment function for cement ball mill according to claim 1, characterized in that: It also includes a data communication module for transmitting wear assessment results, trend data and warning information to a host monitoring platform or remote expert system in real time via industrial Ethernet or wireless communication.

Images