A predictive maintenance system for a spindle of a numerically controlled machine tool

Abstract

The application is a predictive maintenance system for spindle of numerical control machine tool, belonging to the field of mechanical equipment health management and intelligent manufacturing technology, comprising a multi-source data acquisition and preprocessing module, which is used to acquire real-time power of spindle driving motor and multi-channel temperature signals, process the real-time power of spindle driving motor to estimate real-time cutting torque sequence and real-time total power sequence of spindle, and process the multi-channel temperature signals to obtain multi-channel real-time temperature sequence; a basic degradation state evaluation module, which is used to calculate mechanical cumulative damage quantitative index based on the real-time cutting torque sequence, calculate cumulative thermal damage index based on the multi-channel real-time temperature sequence, and fuse the mechanical cumulative damage quantitative index and the cumulative thermal damage index to generate a basic degradation index. The application avoids the risk of missed judgment caused by failure or insensitivity of a single signal source, and significantly improves the robustness and operability of the prediction result in the real industrial environment.

Description

Technical Field

[0001] This invention relates to the field of mechanical equipment health management and intelligent manufacturing technology, specifically a predictive maintenance system for CNC machine tool spindles. Background Technology

[0002] The CNC machine tool spindle is a core power component that determines machining accuracy and efficiency. Its unexpected failure can cause severe production interruptions and economic losses. Therefore, predictive maintenance of the spindle, accurately predicting its remaining service life, is crucial for ensuring production continuity. Existing technologies, when dealing with complex and variable machining conditions, generally suffer from problems such as distorted degradation rate estimation and an inability to effectively distinguish between normal operating condition fluctuations and actual performance degradation. This results in low reliability of the prediction results, making them difficult to apply effectively in flexible manufacturing environments.

[0003] The information disclosed in the background section above is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] The purpose of this invention is to provide a predictive maintenance system for CNC machine tool spindles to solve the problems mentioned in the background art.

[0005] The technical solution of the present invention includes: a multi-source data acquisition and preprocessing module, used to acquire the real-time power of the spindle drive motor and multiple temperature signals, process the real-time power of the spindle drive motor to estimate the real-time cutting torque sequence and the real-time total spindle power sequence, and process the multiple temperature signals to obtain multiple real-time temperature sequences;

[0006] The basic degradation state assessment module is used to calculate the mechanical cumulative damage quantification index based on the real-time cutting torque sequence, calculate the cumulative thermal damage index based on the multi-channel real-time temperature sequence, and fuse the mechanical cumulative damage quantification index and the cumulative thermal damage index to generate the basic degradation index.

[0007] The operating condition variation index calculation module is used to identify power mutation events based on the real-time spindle total power sequence and construct a probability distribution, calculate the information entropy of the probability distribution, normalize the information entropy, and generate the operating condition variation index.

[0008] The adaptive life prediction compensation module is used to determine the basic degradation rate based on the basic degradation index to estimate the basic remaining lifespan, generate prediction confidence and prediction effective time scale based on the operating condition variation index, and combine the basic remaining lifespan, the prediction confidence and the prediction effective time scale to generate a composite life prediction result.

[0009] Preferably, the generation process of the basic degradation index is as follows:

[0010] The mechanical cumulative damage quantification index and the cumulative thermal damage index are normalized.

[0011] The basic degradation index is obtained by weighted fusion of the normalized mechanical cumulative damage quantification index and the cumulative thermal damage index.

[0012] Preferably, the calculation process for the mechanical cumulative damage quantification index is as follows:

[0013] A Fourier transform is performed on the real-time cutting torque sequence within the time window to obtain the power spectral density. The integral of the power spectral density within the preset core frequency band is calculated to obtain an integral result. The integral result is accumulated over time to obtain the mechanical cumulative damage quantification index.

[0014] Preferably, the calculation process for the cumulative thermal damage index is as follows:

[0015] Based on the multi-channel real-time temperature sequence and the preset spatial distance between temperature measurement points, the instantaneous thermal stress characterization quantity is calculated, and the instantaneous thermal stress characterization quantity is accumulated over time to obtain the cumulative thermal damage index.

[0016] Preferably, the process for generating the operating condition variation index is as follows:

[0017] The differential sequence of the real-time spindle total power sequence is calculated, and the amplitude of power mutation events is identified based on an adaptive threshold. Within the analysis time window, the amplitude of the power mutation events is statistically analyzed and an amplitude probability distribution is constructed. Based on the amplitude probability distribution, the power mutation information entropy is calculated, and the power mutation information entropy is normalized to obtain the operating condition variation index.

[0018] Preferably, the process for generating the composite lifetime prediction result is as follows:

[0019] The basic degradation index sequence is fitted to the history to estimate the current degradation rate. Based on the current degradation rate and a preset failure threshold, the basic remaining useful life is calculated. The basic remaining useful life, the prediction confidence, and the prediction effective time scale are combined to generate the composite lifetime prediction result.

[0020] Preferably, the process for generating the prediction confidence level is as follows:

[0021] The predicted confidence level is calculated based on the operating condition variation index and the preset confidence adjustment factor.

[0022] Preferably, the process for generating the effective time scale for prediction is as follows:

[0023] The effective time scale for prediction is obtained by using an exponential decay model based on the operating condition variation index and a preset time scale decay coefficient.

[0024] This invention provides an improved predictive maintenance system for CNC machine tool spindles, which has the following improvements and advantages compared to the prior art:

[0025] 1. By setting up a working condition variation index calculation module, this invention accurately quantifies the complexity of abstract processing working conditions, solving the core pain point of existing technologies that are difficult to distinguish between normal working condition fluctuations and actual performance degradation under varying working conditions. This enables the system to have working condition perception capabilities, clearly distinguishing whether drastic changes in signals are due to normal production scheduling or abnormal performance degradation, thereby greatly improving the accuracy of prediction and effectively suppressing false alarms.

[0026] 2. The basic degradation index constructed in this invention, through the quantification and fusion of two damage sources, mechanical and thermal, achieves a more comprehensive and robust tracking of the spindle health status. The fusion of multi-physics information ensures the comprehensiveness of degradation assessment and avoids the risk of missed judgment caused by the failure or insensitivity of a single signal source.

[0027] 3. The innovative adaptive life prediction compensation module of this invention enables the final composite life prediction result to possess dynamism and practicality, changing the situation of traditional predictive maintenance information being singular and lacking in guidance. When the operating conditions are stable, the system can provide a long-term maintenance planning reference; while when the operating conditions change drastically, the system automatically narrows the prediction range to the short term, prompting the user to observe carefully. This design makes the output of this invention no longer an isolated number, but a context-aware intelligent decision support information containing three elements: predicted value, credibility, and effective time, significantly improving the robustness and operability of the prediction results in real industrial environments. Attached Figure Description

[0028] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0029] Figure 1 This is a flowchart of a predictive maintenance system for CNC machine tool spindles according to the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0031] Example 1:

[0032] Please see Figure 1The present invention provides a predictive maintenance system for CNC machine tool spindles, comprising: a multi-source data acquisition and preprocessing module, used to acquire real-time power of spindle drive motor and multiple temperature signals, process real-time power of spindle drive motor to estimate real-time cutting torque sequence and real-time total spindle power sequence, and process multiple temperature signals to obtain multiple real-time temperature sequences;

[0033] The basic degradation state assessment module is used to calculate the mechanical cumulative damage quantification index based on the real-time cutting torque sequence, calculate the cumulative thermal damage index based on the multi-channel real-time temperature sequence, and integrate the mechanical cumulative damage quantification index and the cumulative thermal damage index to generate the basic degradation index.

[0034] The operating condition variation index calculation module is used to identify power mutation events based on the real-time spindle total power sequence and construct a probability distribution, calculate the information entropy of the probability distribution, normalize the information entropy, and generate the operating condition variation index.

[0035] The adaptive life prediction compensation module is used to determine the basic degradation rate based on the basic degradation index to estimate the basic remaining life, generate prediction confidence and prediction effective time scale based on the operating condition variability index, and combine the basic remaining life, prediction confidence and prediction effective time scale to generate a composite life prediction result.

[0036] In this embodiment, through the precise collaboration of four internal modules, a deep understanding and accurate prediction of the spindle's health status are achieved. The system's input and preprocessing unit, namely the multi-source data acquisition and preprocessing module, continuously captures operational data from key parts of the CNC machine tool spindle, including real-time power of the drive motor. Multiple temperature signals distributed at heat-sensitive locations such as the front and rear bearings of the spindle The core task of this module is not only data acquisition, but also the analysis of the raw signal through a physical model. Based on the principles of electrical engineering, it separates the net cutting power from the total power by eliminating the no-load power. For example, by collecting the power of the spindle when it is not performing cutting operations (i.e., running under no-load), and using this as the no-load power baseline, the net cutting power can be obtained by subtracting this baseline value or the no-load power model value based on the rotational speed from the total power. This allows for the accurate estimation of the real-time cutting torque sequence, which is crucial for assessing mechanical damage. At the same time, it provides the upper-level module with the real-time total spindle power sequence. and multiple real-time temperature sequences These preprocessed, synchronized time-series data form the data foundation of the entire forecasting system.

[0037] Based on the preprocessed data, the basic degradation state assessment module receives cutting torque and temperature data to construct a basic degradation index that can reflect the accumulation of irreversible damage to the spindle. Meanwhile, the operating condition variability index calculation module processes the total power data in parallel, aiming to quantify the complexity and variability of the current processing conditions and generate a dynamic operating condition variability index. The adaptive lifetime prediction and compensation module, as the system's decision-making unit, integrates the outputs of the first two modules: not only based on... Historical trends are used to predict the baseline remaining lifespan, and even more so, historical trends are used to predict the baseline remaining lifespan. The system dynamically compensates for and adjusts the reliability and effective time range of the prediction. This architecture enables the system to output a composite lifetime prediction result that includes the predicted value, confidence interval, and effective time scale. This provides production managers with a much richer, more reliable, and more practically guiding basis for decision-making than traditional methods, thereby improving the effectiveness of predictive maintenance in flexible manufacturing environments.

[0038] Example 2

[0039] The process of generating the basic degradation index is as follows:

[0040] The quantitative indicators of cumulative mechanical damage and cumulative thermal damage were normalized.

[0041] The basic degradation index is obtained by weighted fusion of the normalized mechanical cumulative damage index and the cumulative thermal damage index.

[0042] The calculation process for the quantitative index of cumulative mechanical damage is as follows:

[0043] Fourier transform is performed on the real-time cutting torque sequence within the time window to obtain the power spectral density. The integral of the power spectral density within the preset core frequency band is calculated to obtain an integral result. The integral result is accumulated over time to obtain a quantitative index of mechanical cumulative damage.

[0044] The calculation process for the cumulative thermal damage index is as follows:

[0045] Based on multiple real-time temperature sequences and the preset spatial distance between temperature measurement points, the instantaneous thermal stress characterization quantity is calculated, and the instantaneous thermal stress characterization quantity is accumulated over time to obtain the cumulative thermal damage index.

[0046] In this embodiment, the basic degradation index The underlying logic for its generation lies in the quantification and mathematical integration of two key physical damage mechanisms. To achieve this goal, the basic degradation state assessment module first quantifies mechanical fatigue and thermal stress accumulation in parallel.

[0047] For mechanical damage, quantitative indicators of cumulative mechanical damage The calculation abandons the simple summation of torques, because it cannot reflect the key influence of cyclic stress frequency on fatigue life.

[0048] The technical motivation lies in the fact that fatigue damage in materials is closely related to the frequency and energy distribution of stress cycles. The introduction of power spectral density analysis based on Fourier transform aims to more scientifically measure the accumulated energy of the stress component that contributes the most to damage from a frequency domain perspective. This index is applied at discrete time points. The calculation formula is defined as follows: ;in, Indicates the deadline The mechanical cumulative damage index, with physical dimensions of the square of torque multiplied by time, for example... ; It refers to the index of the discrete time point at which the cutoff occurs; For the j-th time window Internal cutting torque sequence The power spectral density describes the distribution of torque energy with frequency f; It is a preset core frequency band. This frequency band is determined by performing modal analysis or historical data spectrum analysis on a specific spindle to accurately lock the frequency range most relevant to key damage modes such as bearing wear and tool chatter, ensuring the relevance and accuracy of the calculation.

[0049] A feasible method for historical data spectrum analysis is as follows: Collect time series of cutting torque or vibration signals from the spindle during typical heavy-load machining or when known anomalies such as chatter marks or abnormal noises occur; apply a Fast Fourier Transform to this data series to obtain its power spectral density map; identify characteristic frequency bands where the power spectral density value consistently exceeds the background noise of normal operating conditions by more than an order of magnitude under the aforementioned abnormal conditions; for example, the upper and lower limits of the frequency range where 90% of the power spectral peak energy is concentrated can be used as the boundaries of the core frequency band. and By repeating this process on multiple historical failure samples and conducting comprehensive analysis, a universally applicable and robust core frequency band can be determined for a specific model of spindle.

[0050] Using this formula, the basic degradation state assessment module will evaluate the time-varying cutting torque signal. Transformed into a monotonically increasing quantitative indicator The growth slope of this indicator directly reflects the severity of the mechanical load on the main bearing in the recent period, while its cumulative value records the total mechanical fatigue damage of the main bearing since it started operating.

[0051] For thermally induced damage, cumulative thermally induced damage index The calculation aims to quantify the cumulative thermal stress effect caused by spatial temperature inhomogeneity over time. The motivation lies in the fact that thermal damage to the principal shaft is formed by the cumulative effect of the thermal stress field generated by temperature differences between different locations. Therefore, it is necessary to first characterize the instantaneous stress level and then integrate it over time to calculate the instantaneous thermal stress characterization. As a discrete approximation of the space temperature gradient, the formula is: ;in, It is a characterization quantity of instantaneous thermal stress; and It is the first one passed in from module one and the Real-time temperature of each temperature measuring point; This is the known straight-line distance between the two temperature measurement points;

[0052] To effectively capture the thermal gradient generated by the spindle operation, a minimal sensor layout is recommended: at least two temperature sensors should be installed on the outer rings of both the front and rear bearings of the spindle; on each bearing, the two sensors should be arranged radially opposite each other, for example, one at the 12 o'clock position and the other at the 6 o'clock position. This layout not only monitors the overall temperature rise of the bearings but also maximizes the detection of local temperature differences caused by uneven stress or poor lubrication, thus providing a basis for calculating instantaneous thermal stress characterization. Provide high-quality input data;

[0053] Subsequently, the instantaneous quantity is accumulated over time to obtain the cumulative thermal damage index: ;in, That is, the deadline. The cumulative thermal damage index; through these two cascaded calculation steps, the basic degradation state assessment module will output multiple discrete temperature readings. This is transformed into a cumulative thermal damage index that is mathematically consistent with the mechanical damage index. ; It is the j-th discrete time point;

[0054] To generate a unified basic degradation index The modules are fused using a linear weighting method; this process begins with... and Max-min normalization is performed to eliminate dimensional differences and map them to a unified standard. Interval; Perform weighted fusion: ;in, and It is a normalized indicator; weighting coefficient and satisfy Its value can be set based on the experience of experts on a specific machine tool, or determined through data-driven methods such as principal component analysis;

[0055] When using principal component analysis to determine weights, the steps are as follows: obtain normalized cumulative mechanical damage indices that cover at least several complete maintenance cycles. and cumulative thermal damage index Time series data; using these two series as variables, construct a... The input matrix is ​​given by n, where n is the number of time points. Principal component analysis is performed on this matrix to calculate the loading vector of the first principal component, denoted as . These two loading values ​​reflect the contribution of the original variables to the first principal component, i.e., the overall degradation trend of the system; the weighting coefficients are calculated by normalizing the absolute magnitude of the loading values, as shown in the formula: , This method bases the determination of weights on an objective analysis of the inherent correlations within the data itself.

[0056] This fusion step generates a unified, monotonically increasing health indicator. This index comprehensively reflects the cumulative effect of both mechanical and thermal damage, providing a reliable degradation trajectory basis for subsequent lifetime prediction.

[0057] Example 3

[0058] The process of generating the operating condition variability index is as follows:

[0059] The differential sequence of the real-time spindle total power sequence is calculated, and the amplitude of power mutation events is identified based on an adaptive threshold. Within the analysis time window, the amplitude of power mutation events is statistically analyzed and an amplitude probability distribution is constructed. Based on the amplitude probability distribution, the power mutation information entropy is calculated, and the power mutation information entropy is normalized to obtain the operating condition variation index.

[0060] In this embodiment, the operating condition variation index calculation module aims to solve the technical problem that the existing technology cannot distinguish between normal operating condition fluctuations and real performance degradation. By introducing Shannon entropy, a core tool of information theory, the complexity of operating conditions is quantitatively characterized.

[0061] The generation of the operating condition variability index follows a rigorous derivation chain; it is achieved by calculating the real-time total spindle power sequence. difference To capture instantaneous changes in power and employ an adaptive threshold Identify meaningful changes in operating conditions from random noise; this threshold is calculated dynamically, for example, within an analysis window. Internal settings ,in The standard deviation of the power sequence within the window. The value is usually 3, based on the 3-sigma rule in statistics, which ensures the statistical significance of event identification; within a sliding analysis window Within this process, the amplitudes of all identified power surge events are collected, and these amplitudes are divided into N discrete intervals. The event frequency in each interval is then statistically analyzed to construct a probability distribution of the power surge amplitude. The number of intervals N can be based on the data-driven Sturges rule. M is the total number of mutation events within the window, which is adaptively determined; substituting this probability distribution into the definition of Shannon entropy, the power mutation information entropy is calculated. : ;in, The unit is bits, and its value directly reflects the complexity of the working condition; to obtain a standardized dimensionless exponent, the module measures the information entropy. After performing maximum and minimum value normalization, the final operating condition variability index is derived. : ;in, and These are the historical entropy extremes that the system continuously learns and updates during long-term operation, ensuring... The dynamic adaptability of the index;

[0062] Through the above steps, the operating condition variability index calculation module achieves a quantitative representation of operating condition complexity and generates the operating condition variability index. This index provides a key input for the system's adaptive prediction capability, enabling the system to clearly determine whether it is currently in a stable processing phase or a phase of drastic change.

[0063] Example 4

[0064] The process of generating composite lifetime prediction results is as follows:

[0065] The system fits a historical sequence of basic degradation indicators to estimate the current degradation rate. Based on the current degradation rate and a preset failure threshold, it calculates the basic remaining useful life. The basic remaining useful life, prediction confidence, and prediction effective time scale are combined to generate a composite lifetime prediction result.

[0066] The process of generating prediction confidence is as follows:

[0067] The prediction confidence is calculated based on the operating condition variation index and the preset confidence adjustment factor.

[0068] The process of generating the effective time scale for prediction is as follows:

[0069] An exponential decay model is used to calculate the effective time scale for prediction based on the operating condition variation index and the preset time scale decay coefficient.

[0070] In this embodiment, the adaptive lifetime prediction and compensation module generates a composite, intelligent lifetime prediction result. This process begins with an estimation of the baseline remaining useful life. The module then uses a baseline degradation model to evaluate historical baseline degradation indicators. The sequence is fitted to estimate the current degradation rate. ;

[0071] A widely used and effective basic degradation model is the exponential degradation model, mathematically expressed as: Among them, parameters It can be done through historical research. The data sequence was fitted using a nonlinear least squares method. Once the model parameters are determined, the degradation rate at the current time step... This can be achieved analytically by taking the first derivative of the model, i.e. This model is able to describe well the physical processes by which many devices accelerate their degradation as they approach the end of their lifespan;

[0072] Based on this rate calculation, RUL is: ;in, It is a preset failure threshold. This dimensionless parameter is derived from statistical analysis of historical data to determine when the spindle experiences functional failure or its machining accuracy is unacceptable. The critical value that is usually reached;

[0073] Functional failure can be quantified and defined, for example, the total vibration value of the spindle bearing exceeds the alarm limit specified in ISO 10816 for this type of machine, or the surface roughness Ra value of the machined standard specimen exceeds the upper limit required by the drawing three times consecutively; in the historical database, all time points that meet the above failure definition are selected, and the corresponding basic degradation index at that time is recorded. The value; summarizing these failure points The values ​​form a statistical sample set; to establish a threshold that is both sensitive and reliable, the 90th percentile of this sample set can be taken as the failure threshold. This approach ensures that the predictive model is not overly conservative due to extreme cases of early failure.

[0074] this It is not output directly, but is used as a basic prediction value, which is then used by the module to calculate the operating condition variability index. This is compensated and adjusted by generating two key dynamic compensation factors: prediction confidence. and the effective time scale for prediction ;

[0075] Prediction credibility The technical motivation for generating this is that when machine tool operating conditions change drastically, the uncertainty of degradation assessment increases, and the reliability of predictions made under such circumstances is low. Therefore, a mechanism is needed to communicate this uncertainty to the user; the generation formula is... ;in, It represents the prediction confidence level between [0,1] in the final output; It is a confidence adjustment factor greater than 0, used to adjust the sensitivity of confidence. Its value can be determined by backtesting and optimization on historical data containing known false alarms caused by drastic changes in operating conditions, selecting a factor that can effectively suppress false alarms. value;

[0076] The optimization process involves setting an objective function, such as the F1 score, on a historical validation dataset that includes both normal operating conditions and drastic changes in operating conditions. This objective function balances both prediction precision and recall. The optimization is then iteratively adjusted within a preset range, for example, from 0.1 to 2.0, with step sizes of 0.1. For each value of , calculate the F1 score of the system prediction result; select the value that maximizes the F1 score. The value serves as the optimal confidence adjustment factor;

[0077] The reliability of this prediction This enables the system to dynamically adjust the confidence interval of the prediction results, thereby providing effective risk warnings;

[0078] Predicted effective time scale The technical motivation behind generating this value is to overcome the deficiency of traditional prediction systems that output RUL values ​​without time constraints, enabling the system to intelligently adjust its prediction horizon based on the clarity of the current situation. This implementation uses an exponential decay model to construct this relationship: ;in, It is the dynamically adjusted effective time scale for prediction; It is the longest prediction time allowed by the system, which is a priori value set in combination with the factory maintenance cycle and equipment characteristics; This is the time-scale decay coefficient, a dimensionless constant that can be preset according to the severity of the equipment or maintenance strategy. A higher value indicates better performance. The value corresponds to a more conservative forecasting strategy;

[0079] In order to The value setting is more operational and can introduce a factory equipment risk assessment score, which is usually assessed by the maintenance department based on factors such as the importance of the equipment on the production line, maintenance costs and difficulties, and ranges from 1 to 10; The value can be associated with RAS, for example, by setting it using an empirical formula: ,in It is a baseline coefficient (e.g., 2.0), and RAS is the risk score for that device; thus, for devices with high severity (RAS>5), The value will increase accordingly, causing the effective time scale of the forecast to decay faster when the operating conditions change, thereby triggering more timely short-term warnings.

[0080] The effective time scale of the forecast enables the system to provide long-term forecasts under stable operating conditions, while automatically switching to short-term early warning mode when operating conditions fluctuate.

[0081] This module outputs a value based on the base remaining lifetime. Prediction credibility With the effective time scale of prediction The composite prediction results constitute a multi-dimensional and adaptive prediction result, which provides maintenance decisions with more operationally guiding information that includes risk assessment, thereby significantly improving the accuracy and reliability of predictive maintenance.

[0082] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A predictive maintenance system for CNC machine tool spindles, characterized in that, include: The multi-source data acquisition and preprocessing module is used to acquire the real-time power of the spindle drive motor and multiple temperature signals, process the real-time power of the spindle drive motor to estimate the real-time cutting torque sequence and the real-time total spindle power sequence, and process the multiple temperature signals to obtain multiple real-time temperature sequences. The basic degradation state assessment module is used to calculate the mechanical cumulative damage quantification index based on the real-time cutting torque sequence, calculate the cumulative thermal damage index based on the multi-channel real-time temperature sequence, and fuse the mechanical cumulative damage quantification index and the cumulative thermal damage index to generate the basic degradation index. The operating condition variation index calculation module is used to identify power mutation events based on the real-time spindle total power sequence and construct a probability distribution, calculate the information entropy of the probability distribution, normalize the information entropy, and generate the operating condition variation index. An adaptive life prediction compensation module is used to determine the basic degradation rate based on the basic degradation index to estimate the basic remaining lifespan, generate prediction confidence and prediction effective time scale based on the operating condition variation index, and combine the basic remaining lifespan, the prediction confidence and the prediction effective time scale to generate a composite life prediction result. The calculation process for the quantitative index of cumulative mechanical damage is as follows: The real-time cutting torque sequence within the time window is subjected to Fourier transform to obtain the power spectral density. The integral of the power spectral density within the preset core frequency band is calculated to obtain an integral result. The integral result is accumulated over time to obtain the mechanical cumulative damage quantification index. The calculation process for the cumulative thermal damage index is as follows: Based on the multi-channel real-time temperature sequence and the preset spatial distance between temperature measurement points, the instantaneous thermal stress characterization quantity is calculated, and the instantaneous thermal stress characterization quantity is accumulated over time to obtain the cumulative thermal damage index. The process for generating the operating condition variation index is as follows: The differential sequence of the real-time spindle total power sequence is calculated, and the amplitude of power mutation events is identified based on an adaptive threshold. Within the analysis time window, the amplitude of the power mutation events is statistically analyzed and an amplitude probability distribution is constructed. Based on the amplitude probability distribution, the power mutation information entropy is calculated, and the power mutation information entropy is normalized to obtain the operating condition variation index.

2. The predictive maintenance system for CNC machine tool spindles according to claim 1, characterized in that, The process for generating the basic degradation index is as follows: The mechanical cumulative damage quantification index and the cumulative thermal damage index are normalized. The basic degradation index is obtained by weighted fusion of the normalized mechanical cumulative damage quantification index and the cumulative thermal damage index.

3. The predictive maintenance system for CNC machine tool spindles according to claim 1, characterized in that, The process of generating the composite lifetime prediction results is as follows: A sequence of basic degradation indicators composed of historical basic degradation indicators is fitted to estimate the current degradation rate. Based on the current degradation rate and a preset failure threshold, the basic remaining useful life is calculated. The basic remaining useful life, the prediction confidence, and the prediction effective time scale are combined to generate the composite lifetime prediction result.

4. The predictive maintenance system for CNC machine tool spindles according to claim 1, characterized in that, The process of generating the prediction confidence level is as follows: The predicted confidence level is calculated based on the operating condition variation index and the preset confidence adjustment factor.

5. A predictive maintenance system for CNC machine tool spindles according to claim 1, characterized in that, The process of generating the effective time scale for prediction is as follows: The effective time scale for prediction is obtained by using an exponential decay model based on the operating condition variation index and a preset time scale decay coefficient.

Images