Methods for Machine Tool Fault Diagnosis and Remaining Life Prediction under Heterogeneous Working Conditions
By using sensor monitoring and classifying homogeneous working conditions, a one-dimensional reliability model is constructed and high-dimensional fault prediction is performed. This solves the accuracy problem of machine tool fault diagnosis and remaining life prediction under heterogeneous working conditions, and realizes real-time prediction and adaptive improvement.
Patent Information
- Application Number
- CN202310987360.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-08-07
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2043-08-07
AI Technical Summary
Under heterogeneous operating conditions, existing technologies struggle to effectively utilize fault data from homogeneous operating conditions for accurate fault diagnosis and remaining life prediction, leading to prediction bias and poor robustness.
By using sensor-based monitoring methods, homogeneous working conditions are identified and classified, a one-dimensional reliability model is constructed, and this model is applied to high-dimensional fault prediction under heterogeneous working conditions. Combined with real-time working condition identification and historical information updates, real-time prediction of the failure probability and remaining life of machine tool tools is achieved.
It enables real-time prediction of machine tool failure probability and remaining life under heterogeneous working conditions, improves the adaptability and accuracy of prediction, and can effectively integrate the variability of future working conditions.
Smart Images

Figure CN117034757B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of machine tool fault diagnosis and remaining life prediction technology, and in particular to a method for machine tool fault diagnosis and remaining life prediction under heterogeneous working conditions. Background Technology
[0002] Accurate fault diagnosis and remaining tool life prediction are crucial for timely tool replacement and improved manufacturing system productivity. However, the uncertainty and heterogeneity of operating conditions, such as external environment and operating parameters, significantly increase the difficulty of tool fault and remaining tool life prediction. Current models have strong predictive capabilities for machine tool fault diagnosis and remaining tool life prediction under given operating conditions, but their flexibility is poor under dynamic and heterogeneous operating conditions. Due to the non-uniformity of operating conditions caused by variations in workpiece material and cutting parameters, the degradation process of tools also differs at different machining stages. Therefore, it is unreasonable to use a single distribution model to predict tool faults and remaining tool life under different operating conditions.
[0003] Heterogeneity is widespread in engineering systems and can be categorized into variability among individuals, variability under time-varying operating conditions, and variability in system tasks and loads. Studies have shown that failure and remaining life predictions that do not consider heterogeneity will suffer from significant biases. In the reliability assessment of heterogeneous populations, population heterogeneity is usually considered to be caused by the mixing of different homogeneous subpopulations, where subpopulations exhibit different characteristics. For fault diagnosis in heterogeneous situations, related research has proposed dynamic linear degradation models that do not evolve over time and stochastic drift parameter Wiener process models that address non-uniform degradation. In addition, existing research has also studied the reliability problems of certain specific engineering objects in heterogeneous environments, such as mining equipment and dump trucks.
[0004] The aforementioned studies lack classification analysis of fault data under heterogeneous operating conditions. Instead, they use one-dimensional models to estimate parameters and predict faults under mixed operating conditions. However, fault data under heterogeneous operating conditions may differ significantly from data under homogeneous operating conditions. If the tool degradation model under a single homogeneous operating condition is considered a one-dimensional model, then the tool degradation model under mixed operating conditions must be a mixture of multiple one-dimensional models. Therefore, the assumption of independent and identically distributed fault data under heterogeneous operating conditions no longer applies. Consequently, using a single one-dimensional model to fit fault data under heterogeneous operating conditions will significantly increase prediction bias. On the other hand, considering the uncertainty of future operating conditions, neglecting the variability of future operating conditions in remaining life prediction will also lead to poor robustness of the prediction results.
[0005] The fundamental task of the CNC machine tool tool prediction module is to identify the current and future failure probabilities of the machine tool. Therefore, in fault diagnosis and remaining life prediction, in addition to considering the cumulative failure risk up to the current monitoring time, the impact of future degradation path variability should also be considered. Related research has also explored different aspects of heterogeneity. For example, the consideration of inter-unit heterogeneity in remaining life prediction methods based on the Wiener process model, the consideration of different lifespan and degradation data in reliability assessment, and the consideration of component heterogeneity under time-varying conditions in the two-factor state-space model for predicting the remaining life of mechanical devices. However, current research still lacks effective utilization of data under homogeneous operating conditions. Machines operate under normal conditions for most of their service life, with few failures, making fault and remaining life prediction a small-sample prediction problem. This further increases the difficulty of making relevant predictions based solely on fault data under heterogeneous operating conditions. Furthermore, even if a model can be built to predict machine tool failure behavior using fault data under given heterogeneous operating conditions, the applicability of such a model under other different heterogeneous operating conditions remains uncertain.
[0006] Sensor monitoring is an effective technology for tracking the real-time status of machine tools, identifying potential operating states by extracting spatial, temporal, and statistical features from large-scale data. This functionality relies heavily on advanced identification algorithms. Examples include statistical time-frequency analysis methods for detecting structural changes in rotating machinery, dynamic weighted principal component analysis algorithms and hierarchical monitoring strategies for detecting fault types, and hybrid methods combining manual feature design and automatic feature learning for machine health identification. Current research primarily focuses on identifying fault states from a large number of healthy states, without mining the useful information contained within normal operating conditions. Even when a machine tool is fault-free, different operating conditions can lead to different degradation processes, which is crucial for fault diagnosis and remaining life prediction. Summary of the Invention
[0007] Predictive models obtained by estimating parameters using a single model for fault data under heterogeneous operating conditions often lack adaptability to fault and remaining life prediction under dynamic operating conditions, increasing the risk of prediction deviation under uncertain operating conditions. This invention proposes a sensor-based method for fault diagnosis and remaining life prediction of machine tool tools under heterogeneous operating conditions. First, it fully mines machine tool operating information under homogeneous and different operating conditions to construct a one-dimensional prediction model of machine reliability under homogeneous operating conditions. Second, it uses the one-dimensional prediction model under homogeneous operating conditions as prior knowledge to assist in the prediction of faults or remaining life of machine tool tools under heterogeneous operating conditions, thus realizing real-time prediction of the remaining life of machine tools under heterogeneous operating conditions.
[0008] The technical solution of this invention is implemented as follows:
[0009] A method for fault diagnosis and remaining life prediction of machine tool cutting tools under heterogeneous working conditions, the steps of which are as follows:
[0010] Step 1: Based on the homogeneous working condition classification method, conduct experiments on the machine operation process under different working conditions and collect the failure time data of machine tool tools under different working conditions;
[0011] Step 2: Based on the collected machine tool failure time data, construct a one-dimensional homogeneous reliability model of the machine tool using the Weibull model;
[0012] Step 3: Identify heterogeneous operating conditions during machine tool operation based on real-time monitoring data from sensors;
[0013] Step 4: Under heterogeneous working conditions, the one-dimensional homogeneous reliability model is mixed to establish a high-dimensional fault prediction model for machine tool tools under heterogeneous working conditions, and the real-time fault probability prediction is performed by combining the real-time heterogeneous working condition identification results.
[0014] Step 5: Based on the real-time monitoring and updating of historical working condition information by sensors, the high-dimensional fault prediction model is updated according to the real-time updated historical working condition information. Based on the updated high-dimensional fault prediction model, the remaining life of machine tool tools is predicted.
[0015] The homogeneous operating condition classification method includes:
[0016] In machine tools, the heterogeneity of operating conditions is represented by the following vector:
[0017] V F = [A, B];
[0018] Where A is 1×n d An n-dimensional vector represents a set of working condition variables with discrete properties. d B represents the number of discrete attribute condition variables; B is a 1×n c An n-dimensional vector represents a set of operating condition variables with continuous attributes. c It is the number of continuous attribute operating condition variables; V F A set representing operating condition variables;
[0019] The criteria for determining whether two operating conditions are identical are as follows:
[0020] {c pi and c pj Homogeneous || A i -A j |=0&|B′ i -B′ j |≤ε};
[0021] Among them, A i Indicates operating condition c piThe values of the discrete variable A are given below. j Indicates operating condition c pj The discrete variable B′ takes values. i For operating condition c pi The normalized value of the continuous attribute variable, B′ j For operating condition c pj The normalized value of the continuous attribute variable, |A i -A j | and |B′ i -B′ j | are all Euclidean vector distances, and ε≥0 is the threshold for judging homogeneous working conditions; c pi and c pj These represent two operating conditions at different time periods.
[0022] The method for constructing a one-dimensional homogeneous reliability model of machine tool cutting tools using the Weibull model is as follows:
[0023] For the operation of a machine tool, all defined homogeneous working conditions constitute its possible set of working conditions, denoted as . Where, n h This represents the total number of all possible homogeneous operating conditions in the machine tool. By performing independent model parameter estimation on the fault data under each homogeneous operating condition, a one-dimensional prediction model R for machine tool faults under homogeneous operating conditions is obtained. b ={R bi (t), i = 1, 2, ..., n h};
[0024] After the machine tool has been running under the same operating conditions for any time t, the failure probability and failure rate of the cutting tool are denoted as follows: and
[0025] In step three, the specific implementation method is as follows:
[0026] Given the homogeneous working condition c pi The machine tool's reference sensor data is as follows:
[0027]
[0028] Where k is the number of sensors, d 0_ij =[ds 0_ij1 ;ds 0_ij2 ;ds 0_ij3 ...] represents a given homogeneous working condition c pi The data sample set obtained by the j-th sensor at different times, j = 1, 2, 3, ..., k, ds 0_ijl Let the sensor data sample vector obtained by the j-th sensor at time l with a specific sampling frequency be represented as follows after normalizing its amplitude:
[0029] ds 0_ijl =ds 0_ijl / max(ds 0_ijl );
[0030] During machine tool operation, based on monitoring data from k sensors, the real-time normalized monitoring data sample is obtained as follows:
[0031] D = {ds1, ds2, ..., ds} k};
[0032] By comparing the similarity between sensor monitoring data D and reference sensor data D0, the current operating condition of the machine tool can be identified; the reference sensor sample ds 0_ijl Real-time monitoring samples ds of sensor j j They are labeled as data 1 and 2 respectively;
[0033] In the amplitude domain, the area enclosed by the probability density function curves of the two data samples is defined as the deviation D. Y This is used to estimate their similarity in the amplitude domain, and its calculation formula is:
[0034]
[0035] Among them, f Y1 (y) is the probability density function of the normalized amplitude of the reference sensor data sample, f Y2 (y) is the probability density function of the normalized amplitude of the real-time monitoring sensor data sample; y is the normalized amplitude; since D Y If ∈[0,2], then the deviation of the normalization is:
[0036]
[0037] In the time domain, the cross-correlation coefficient can be used to measure the similarity between two time series, and its expression is:
[0038]
[0039] Among them, ds 0_ijl (t) is the reference sensor data sample ds 0_ijl The time series, denoted as time series 1; ds j (t+τ) is the real-time monitored sensor data sample ds j The time series with a delay of τ, denoted as time series 2, has its normalized cross-correlation coefficient expressed as:
[0040]
[0041] Among them, R11 (0) is the reference sensor data sample ds 0_ij1 The autocorrelation coefficient, R 22 (0) is the real-time sensor monitoring data sample ds j The autocorrelation coefficient; the maximum normalized cross-correlation coefficient obtained for different delay times τ. Represented as This can represent the best level of similarity between the two data samples;
[0042] Based on the definitions of deviation and cross-correlation coefficient, the similarity between two sensor data samples is inversely proportional to the deviation and directly proportional to the cross-correlation coefficient. Therefore, the average difference between the cross-correlation coefficient and the deviation of k monitoring sensors is used to measure the similarity of sensor monitoring data under two operating conditions, i.e., to measure the similarity between the two operating conditions. The calculation formula is as follows:
[0043]
[0044] In addition, deviation If both the cross-correlation coefficient and the cross-correlation coefficient are normalized values, meaning they fall within the range [0,1], then the two points... The Euclidean distance between (0,1) and (0,1) is used to measure the similarity of sensor monitoring data under two operating conditions, i.e., to measure the similarity between the two operating conditions. The calculation formula is:
[0045]
[0046] Among them, SL dif SL represents a similarity index calculated based on differences. dis This represents a similarity index calculated based on Euclidean distance, where n k It is the number of sensors used for similarity level assessment. It is a similarity evaluation standard based on the amplitude range. It is a time-domain based similarity evaluation standard; for a given similarity index SL, SL = SL dis Or SL = SL dif By comparing real-time monitored sensor data samples with a certain homogeneous operating condition c pi All reference sensor data samples D 0i By comparing the current real-time monitoring data sample with that of the same operating condition c, we can find the data. pi The level of most similarity is denoted as SL. i This indicates the current operating condition and the condition of component c under the same operating conditions. pi The similarity level between them; by comparing real-time monitoring data with all homogeneous operating conditions c piBy comparing the real-time monitoring data with all similar operating conditions, the similarity index value can be obtained. The following formula can then be used to determine which type of similar operating condition the processing stage containing the real-time monitoring data belongs to:
[0047]
[0048] in, This indicates that the current real-time monitoring data has been determined to be in the same type of working condition.
[0049] The high-dimensional fault prediction model for machine tool cutting tools under heterogeneous working conditions is as follows:
[0050] F(t)=f(R b (t),N b (t),T b (t));
[0051] Where F(t) is the high-dimensional fault prediction model, f(·) represents the functional expression of the high-dimensional fault prediction model, and N b (t) represents the historical operating condition sequence up to the current time t, where T is the operating condition sequence. b (t) represents the duration of the corresponding operating condition sequence up to the current time t.
[0052] The method for real-time fault probability prediction includes:
[0053] Using machine tool failure probability prediction under homogeneous operating conditions as prior knowledge, and based on the one-dimensional failure rate model under homogeneous operating conditions, a high-dimensional weighted failure rate model under heterogeneous operating conditions can be constructed:
[0054]
[0055] Where, τ i (t) represents the machine tool operating under the same conditions up to time t. pi The proportion of time spent working, and At any time t, under any homogeneous working condition c pi The cumulative time ratio τ i (t) can be represented as:
[0056]
[0057] Among them, t c This is the starting point of the current homogeneous operating condition, a i This indicates the homogeneous operating condition c before the current homogeneous operating condition begins. pi The cumulative working hours, b i Indicates whether the current operating condition is a homogeneous operating condition (c). pi Judgment coefficient: b i =1 indicates that the current working condition is the same as condition c. pi Otherwise bi =0;
[0058] According to the definition of failure rate, the following differential equation exists:
[0059]
[0060] Among them, f m F(t) is the probability density function of machine tool failure under heterogeneous operating conditions. m (t) is the cumulative distribution function of machine tool faults under heterogeneous operating conditions; considering F m (0) = 0 and r m (0) = 0, and the actual cumulative distribution function of the machine tool under heterogeneous working conditions is:
[0061]
[0062] For any operating time interval [t1, t2] under homogeneous operating conditions, there are two possible cases:
[0063] (1) If b i If = 0, then the formula for calculating the actual cumulative distribution function can be obtained.
[0064]
[0065] Where, α ri Let λ represent the shape parameter of the one-dimensional reliability model of machine tool cutting tools under the Weibull distribution. ri This represents the scale parameters of a one-dimensional reliability model for machine tools under the Weibull distribution;
[0066] (2) If b i =1, and the integral can be calculated using the formula described above.
[0067]
[0068] Based on the above two scenarios, the integral formula for calculating the actual cumulative distribution function of the machine tool under heterogeneous working conditions can be summarized as follows:
[0069]
[0070] The integration range is [0,t1], [t1,t2], ..., [t c [t] represents the continuous operating time of the machine tool under a certain homogeneous working condition, k = 1, 2, 3, ... K is the duration sequence, and K is the cumulative number of occurrences of all different homogeneous working conditions; furthermore, t0 = 0, t K-1 =t c , and t K =t.
[0071] In step five, the specific implementation method is as follows:
[0072] At any given time t, obtain real-time updated historical operating condition information, including operating condition sequence N. b (t) and its duration sequence T b (t);
[0073] Based on the operating condition sequence, obtain the state transition matrix P between homogeneous operating conditions of the machine tool:
[0074]
[0075] Where, p ij The machine tool is in working condition c pi to c pj The transition probability;
[0076] Based on the latest state transition matrix, the transition trajectory N′ of the future operating condition is obtained through simulation. b ;
[0077] Based on real-time updated historical operating condition information, the time distribution characteristics of each homogeneous operating condition are updated in real time, and then the duration T′ of different future operating conditions is simulated to obtain the result. b ;
[0078] Based on the Monte Carlo method, the possible life trajectory of machine tool tools is estimated, and the remaining tool life and its distribution function are predicted and analyzed by simulating the future working condition sequence of the machine tool.
[0079] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0080] 1) This invention proposes a classification and identification method for homogeneous and heterogeneous working conditions based on sensor monitoring, and constructs a fault diagnosis model based on working condition identification, realizing real-time prediction of tool failure probability under changing working conditions.
[0081] 2) An online prediction method for remaining tool life based on historical working condition information is proposed. Through this method, the life variability caused by the heterogeneity of future working conditions can be well integrated into the prediction model, and the remaining tool life can be predicted in real time. Attached Figure Description
[0082] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0083] Figure 1 This is a flowchart of the present invention.
[0084] Figure 2 This is a framework diagram of the machine tool fault prediction based on sensor monitoring according to the present invention.
[0085] Figure 3 This is a similarity estimate for signal samples in the amplitude domain.
[0086] Figure 4 This is a flowchart of the online remaining useful life prediction process based on historical operating conditions.
[0087] Figure 5 The curves show the failure probability of machine tool cutting tools under different homogeneous working conditions.
[0088] Figure 6 Statistical results for identifying the optimal operating condition.
[0089] Figure 7 The curves represent fault prediction and remaining life prediction based on operating condition identification.
[0090] Figure 8 This is a histogram of failure probabilities for mode M0. Detailed Implementation
[0091] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0092] like Figure 1 As shown, this embodiment of the invention provides a method for fault diagnosis and remaining life prediction of machine tool tools under heterogeneous operating conditions based on sensor monitoring. It utilizes multiple reliability models based on fault data from homogeneous operating conditions to construct a composite prediction model for heterogeneous operating conditions, enabling adaptive prediction of machine tool fault probability and remaining life under heterogeneous operating conditions. The specific steps are as follows:
[0093] Step 1: Based on the homogeneous working condition classification method, conduct experiments on the machine operation process under different working conditions and collect machine tool failure time data under different working conditions.
[0094] The homogeneous operating condition classification method includes:
[0095] For a machine tool, the external environment and operating parameters determine whether its operating conditions are heterogeneous. Variations in the former are typically caused by input work-in-process, including their material properties and product quality characteristics. The latter usually consists of various cutting parameters of the machine tool, such as depth of cut (DOC), feed rate, and cutting speed. In a machine tool, operating condition heterogeneity is represented by the following vector:
[0096] V F = [A, B];
[0097] Where A is 1×n d A dimensional vector representing a set of working condition variables with discrete properties, such as material type; n d B represents the number of discrete attribute condition variables; B is a 1×n c A dimensional vector represents a set of working condition variables with continuous attributes, such as workpiece dimensions, DOC, feed rate, and cutting speed; n c It is the number of continuous attribute operating condition variables; V F This represents a set of operating condition variables.
[0098] Having the same discrete vector A is a prerequisite for homogeneity of operating conditions. However, small changes in the continuous vector B do not always produce significant differences between operating conditions. Therefore, a criterion is needed to identify potentially homogeneous operating conditions, as follows:
[0099] {c pi and c pj Heterogeneous || A i -A j |=0&|B′ i -B′ j |≤ε};
[0100] Among them, A i Indicates operating condition c pi The values of the discrete variable A are given below. j Indicates operating condition c pj The values of the discrete variables under B′ i For operating condition c pi The normalized value of the continuous attribute variable, B′ j For operating condition c pj The normalized value of the continuous attribute variable, |A i -A j | and |B′ i -B′ j | are all Euclidean vector distances, and ε≥0 is the threshold for judging homogeneous working conditions; c pi and c pj These represent two operating conditions at different time periods.
[0101] Step 2: Based on the collected machine tool failure time data, construct a one-dimensional homogeneous reliability model of the machine tool using the Weibull model.
[0102] The consistency and completeness of prior knowledge are crucial for the accuracy and effectiveness of subsequent predictions, which can be ensured through local testing under homogeneous operating conditions. Analysis of fault data and parameter estimation are fundamental methods for measuring the failure risk of engineering objects. Under given homogeneous operating conditions, the failure probability distribution obtained from fault data analysis can predict the failure probability of the engineering object. The Weibull distribution has good flexibility in describing different failure modes, and can describe different failure modes with decreasing, stable, and increasing failure probabilities. Therefore, this invention selects the Weibull model to construct a one-dimensional failure prediction model for machine tool cutting tools under homogeneous operating conditions. For a given Weibull distribution model, its probability density function (PDF), cumulative distribution function (CDF), and failure rate are as follows:
[0103]
[0104]
[0105]
[0106] Where, α r >0、λ r >0 represents the shape parameter and the dimensional parameter, respectively, which can be estimated from the fault data. For the operation of a machine tool, all defined homogeneous working conditions constitute its possible set of working conditions, denoted as . Where, n h It represents the total number of all possible homogeneous operating conditions in the machine tool; by using fault data under each homogeneous operating condition, independent model parameter estimation is performed to obtain a one-dimensional prediction model R for machine tool faults under homogeneous operating conditions. b ={R bi (t), i = 1, 2, ..., n h}; Ensure the accuracy of the prediction model under a single, homogeneous working condition.
[0107] Assuming the machine tool operates under homogeneous conditions, after any time t, the failure probability and failure rate of the tool are denoted as follows: and These are defined as the basic failure probability and the basic failure rate, respectively. Based on the one-dimensional fault prediction model under the above homogeneous operating conditions, prior knowledge can be provided for fault diagnosis and remaining life prediction of machine tools under heterogeneous operating conditions.
[0108] Step 3: Identify heterogeneous operating conditions during machine tool operation based on real-time monitoring data from sensors.
[0109] Heterogeneous operating conditions generate heterogeneous failure data for machine tools, such as mean time between failures (MTBF). Therefore, the assumption that failure data under heterogeneous conditions are independent and identically distributed does not apply, and the risk analysis results for heterogeneous data may differ significantly from those under homogeneous conditions. To improve the predictive capability of machine tools under heterogeneous operating conditions, this invention proposes a two-step prediction method based on sensor monitoring, such as... Figure 2 As shown.
[0110] In model building, the lifespan data of machine tool cutting tools under different homogeneous working conditions is first collected based on experimental analysis. Then, a one-dimensional homogeneous reliability model R for the machine tool cutting tools is constructed based on the collected data. b ={R bi (t), i = 1, 2, ..., n h}, where n h This represents the total number of all possible homogeneous operating conditions in the machine tool. Then, the corresponding one-dimensional homogeneous reliability models are combined to establish a high-dimensional fault prediction model F(t) for machine tool tools under heterogeneous operating conditions:
[0111] F(t)=f(R b (t),N b (t),T b (t));
[0112] Where f(·) represents the functional expression of the high-dimensional fault prediction model, and N b (t) represents the historical operating condition sequence up to the current time t, where T is the operating condition sequence. b (t) represents the duration of the corresponding operating condition sequence up to the current time t.
[0113] Secondly, in the sensor-based prediction process, the occurrence of heterogeneous operating conditions of the machine tool is first monitored and identified using sensors. Through sensors configured on the machine tool, different homogeneous operating conditions can be identified, resulting in the occurrence sequence N of these homogeneous operating conditions. b (t) and its duration T b (t). Based on the above occurrence sequence and duration, we can obtain c for each homogeneous working condition. pi Frequency of occurrence N bi (t) and cumulative duration T bi (t). Therefore, based on the above dynamic parameters, the failure probability and remaining life of machine tool tools can be predicted in real time, and the variability of machine tool tool reliability caused by the heterogeneity of working conditions can be integrated into the prediction model.
[0114] When a machine tool operates under multiple homogeneous operating conditions, heterogeneous operating conditions are generated. According to a sensor-based prediction framework, real-time identification of heterogeneous operating conditions through monitoring data is a necessary condition for prediction. Given a homogeneous operating condition c... pi The machine tool's reference sensor data is as follows:
[0115]
[0116] Where k is the number of sensors, d 0_ij =[ds 0_ij1 ;ds 0_ij2 ;ds 0_ij3 ...] represents a given homogeneous working condition c pi The data sample set obtained by the j-th sensor at different times, j = 1, 2, 3, ..., k, ds 0_ijl Let the sensor data sample vector obtained by the j-th sensor at time l with a specific sampling frequency be represented as follows after normalizing its amplitude:
[0117] ds 0_ijl =ds 0_ijl / max(ds 0_ijl );
[0118] During machine tool operation, based on monitoring data from k sensors, the real-time normalized monitoring data sample is obtained as follows:
[0119] D = {ds1, ds2, ..., ds} k};
[0120] By comparing the similarity between sensor monitoring data D and reference sensor data D0, the current operating condition of the machine tool can be identified; the reference sensor sample ds 0_ijl Real-time monitoring samples ds of sensor j j They are labeled as data 1 and 2 respectively; the results are as follows Figure 3 As shown.
[0121] In the amplitude domain, the area enclosed by the probability density function curves of the two data samples is defined as the deviation D. Y This is used to estimate their similarity in the amplitude domain, such as Figure 3 As shown, the calculation formula is:
[0122]
[0123] Among them, f Y1 (y) is the probability density function of the normalized amplitude of the reference sensor data sample, f Y2 (y) is the probability density function of the normalized amplitude of the real-time monitoring sensor data sample; y is the normalized amplitude; since D Y If ∈[0,2], then the deviation of the normalization is:
[0124]
[0125] In the time domain, the cross-correlation coefficient is used to measure the similarity between two time series, and its expression is:
[0126]
[0127] Among them, ds 0_ijl (t) is the reference sensor data sample ds 0_ijl The time series, denoted as time series 1; ds j (t+τ) is the real-time monitored sensor data sample ds j Let the time series with a delay of τ be denoted as time series 2. Its normalized cross-correlation coefficient can be expressed as:
[0128]
[0129] Among them, R 11 (0) is the sensor data reference sample ds 0_ij1 The autocorrelation coefficient, R 22 (0) is the sensor data monitoring sample ds j The autocorrelation coefficient; the maximum normalized cross-correlation coefficient obtained for different delay times τ. Represented as This represents the best level of similarity between the two data samples.
[0130] Based on the definitions of deviation and cross-correlation coefficient, the similarity between two sensor data samples is inversely proportional to the deviation and directly proportional to the cross-correlation coefficient. Therefore, the average difference between the cross-correlation coefficient and the deviation of k monitoring sensors can be used to measure the similarity of sensor monitoring data under two operating conditions, i.e., to measure the similarity between two operating conditions. The calculation formula is as follows:
[0131]
[0132] In addition, the deviation If both the cross-correlation coefficients and the cross-correlation coefficients are normalized values, i.e., falling between [0,1], then the two points... The Euclidean distance between (0,1) and (0,1) can also measure the similarity of sensor monitoring data under two operating conditions, i.e., the degree of similarity between the two operating conditions. The calculation formula is:
[0133]
[0134] Among them, SL dif SL represents a similarity index calculated based on differences. dis This represents a similarity index calculated based on Euclidean distance, where n k It is the number of sensor types used for similarity level assessment. It is a similarity evaluation standard based on the amplitude range. It is a time-domain similarity evaluation standard; for a given similarity index SL (SL = SL... dis Or SL = SL dif By comparing real-time monitored sensor data samples with a certain homogeneous operating condition c pi All reference sensor data samples D 0i By comparing the current real-time monitoring data sample with that of the same operating condition c, we can find the data. pi The level of most similarity is denoted as SL. i This indicates the current operating condition and the condition of component c under the same operating conditions. pi The similarity level between them; by comparing real-time monitoring data with all homogeneous operating conditions c pi By comparing the real-time monitoring data with all similar operating conditions, the similarity index value can be obtained. The following formula can then be used to determine which type of similar operating condition the processing stage described in the real-time monitoring data belongs to:
[0135]
[0136] in, This indicates that the current real-time monitoring data has been determined to be in the same type of working condition.
[0137] Step 4: Under heterogeneous working conditions, the one-dimensional homogeneous reliability model is mixed to establish a high-dimensional fault prediction model for machine tool cutting tools under heterogeneous working conditions, and the real-time fault probability prediction is performed by combining the real-time heterogeneous working condition identification results.
[0138] Under heterogeneous operating conditions, each segment of homogeneous operating conditions leads to the accumulation of machine tool failure risks. Using the prediction of machine tool failure probability under homogeneous operating conditions as prior knowledge, and based on the one-dimensional basic failure rate model under homogeneous operating conditions, a high-dimensional weighted failure rate model under heterogeneous operating conditions can be constructed:
[0139]
[0140] Where, τ i (t) represents the machine tool operating under the same conditions up to time t. pi The proportion of time spent working, and At any time t, under any homogeneous working condition c pi The cumulative time ratio τ i (t) can be represented as:
[0141]
[0142] Among them, t c This is the starting point of the current homogeneous operating condition, a i This indicates the homogeneous operating condition c before the current homogeneous operating condition begins. pi The cumulative working hours, b iIndicates whether the current operating condition is a homogeneous operating condition (c). pi Judgment coefficient: b i =1 indicates that the current working condition is the same as condition c. pi Otherwise b i =0. Furthermore, τ is defined as... i (0) = 0. Time ratio τ i (t) determines the high-dimensional failure rate under heterogeneous conditions and a certain homogeneous condition c. pi The degree of deviation in the next dimension failure rate. In the case of only a single heterogeneous operating condition c... pi In extreme cases, τ i (t) = 1.
[0143] According to the definition of failure rate, the following differential equation exists:
[0144]
[0145] Among them, f m F(t) is the probability density function of machine tool failure under heterogeneous operating conditions. m (t) is the cumulative distribution function of machine tool faults under heterogeneous operating conditions; considering F m (0) = 0 and r m (0) = 0, and the actual cumulative distribution function of the machine tool under heterogeneous working conditions is:
[0146]
[0147] For any operating time interval [t1, t2] under homogeneous operating conditions, there are two possible cases:
[0148] (1) If b i If = 0, then the formula for calculating the actual cumulative distribution function can be obtained.
[0149]
[0150] Among them, a ri Let λ represent the shape parameter of the one-dimensional reliability model of machine tool cutting tools under the Weibull distribution. ri This represents the scale parameter of the one-dimensional reliability model of machine tool cutting tools under the Weibull distribution.
[0151] (2) If b i =1, and the integral can be calculated using the formula described above.
[0152]
[0153] Based on the above two scenarios, the integral formula for calculating the actual cumulative distribution function of the machine tool under heterogeneous working conditions can be summarized as follows:
[0154]
[0155] The integration range is [0,t1], [t1,t2], ..., [t c [t] represents the continuous operating time of the machine tool under a certain homogeneous working condition, k = 1, 2, 3, ... K is the duration sequence, and K is the cumulative number of occurrences of all different homogeneous working conditions; furthermore, t0 = 0, t K-1 =t c , and t K =t.
[0156] Step 5: Based on the real-time monitoring and updating of historical working condition information by sensors, the high-dimensional fault prediction model is updated according to the real-time updated historical working condition information. Based on the updated high-dimensional fault prediction model, the remaining life of machine tool tools is predicted.
[0157] Within a sensor-based prediction framework, this embodiment proposes a dynamic prediction method for remaining lifetime based on real-time historical operating condition information, such as... Figure 4 As shown. At any current time t, obtain real-time updated historical operating condition information, including operating condition sequence N. b (t) and its duration sequence T b (t). Then, based on the working condition sequence, the state transition matrix P between homogeneous working conditions of the machine tool is obtained:
[0158]
[0159] Where, p ij The machine tool is in working condition c pi to c pj The transition probability is synchronously updated in real time with dynamic historical operating condition sequence information. Based on the latest state transition matrix, the transition trajectory N′ of the future operating condition is obtained through simulation. b Furthermore, based on real-time updated historical operating condition information, the time distribution characteristics of each homogeneous operating condition are updated in real time, thereby simulating the duration T′ of different future operating conditions. b Finally, based on the Monte Carlo method, the possible life trajectory of the machine tool is estimated, and the remaining tool life and its distribution function are predicted and analyzed by simulating the future operating condition sequence of the machine tool.
[0160] Application Analysis:
[0161] The NASA Ames Prediction Database is a continuously updated database encompassing multiple sets of prediction data provided to universities, companies, and institutions. It provides datasets of milling cutter flank wear under various operating conditions. This example uses a degradation dataset generated by a Matsuura MC-510V machining center, where the workpiece dimensions are 483mm × 178mm × 51mm. The discrete attribute variables for the machining conditions are material type, including cast iron and J45 stainless steel; the continuous attribute variables include DOC (1.50 mm and 0.75 mm) and feed rate (0.5 mm / revolution and 0.25 mm / revolution), with a continuous attribute threshold of ε = 0. In summary, this example considers three operating condition variables (material, DOC, and feed rate). As shown in Table 1, the three operating condition variables constitute n... h =2 3 Eight homogeneous working conditions were used, and two cutting experiments were conducted for each condition, resulting in a total of 16 experiments. A 70mm face milling cutter with six inserts was selected as the cutting tool, and the flank wear width (VB) was used to describe tool life. Additionally, k=6 sensors were used to measure AC and DC spindle motor current signals (smsAC, smsDC), table and spindle vibration signals (vib_table, vib_spindle), and table and spindle acoustic signals (AE_table, AE_spindle), for a total of six types of sensor signals. Finally, 165 sets of valid degradation and sensor data samples were obtained.
[0162] Table 1. Results of the study under the same working conditions in the experiment.
[0163]
[0164] When the failure threshold for end mill flank wear is set to 4.00, the Weibull distribution failure model parameters for the tool under each homogeneous working condition can be estimated, as shown in Table 2. The corresponding failure probability curves of the machine tool under different homogeneous conditions are as follows: Figure 5 As shown.
[0165] Table 2. Failure distribution parameters under different homogeneous conditions
[0166]
[0167] Based on 165 sets of valid degradation and sensor data samples, this example analyzes the effectiveness of the proposed operating condition identification method. Using the identification method mentioned in Section 4.1, each set of sensor data is used as a real-time sensor data sample and compared with other sets to test whether the same operating condition is identified.
[0168] Figure 6The statistical analysis results of the proposed method for identifying variables under different operating conditions are presented. Furthermore, based on the 11 time-domain characteristic parameters (TDF1~TDF) in Table 3... 11 ) and the 13 frequency domain characteristic parameters (FDF) in Table 4 12 ~FDF 24 The method uses feature parameter-based similarity recognition as a comparison method, taking the minimum difference in feature parameters between real-time data samples and benchmark samples as the recognition criterion, and performs recognition analysis, and compares it with the method proposed in this invention.
[0169] Table 3 Time-domain characteristics
[0170]
[0171] Table 4 Frequency Domain Characteristics
[0172]
[0173]
[0174] Recognition results as follows Figure 6 As shown, for different operating condition variables, the SL evaluation criterion proposed in this example achieves superior recognition performance compared to traditional feature parameter evaluation methods. Furthermore, this example also obtains the optimal sensor type and SL evaluation criterion for identifying operating condition variables under different recognition methods, as shown in Table 5. The results indicate that operating condition variables have different sensitivities to sensor data types, leading to differences in the recognition capabilities of different sensor data combinations for different operating condition variables. In the proposed method, the Euclidean distance index SL... dis Comparison Difference Index SL dif It has a stronger ability to identify working conditions.
[0175] Table 5. Optimal sensor data types and criteria for different operating conditions under different methods
[0176]
[0177] Based on the heterogeneous operating conditions composed of the eight homogeneous operating conditions in Table 1, assume that the operating condition state transition matrix obtained at a certain moment is as shown in the following equation. At the current time t, fault probability diagnosis and remaining life prediction can be achieved.
[0178]
[0179] Assuming the duration of each homogeneous operating condition at the current time t follows a normal distribution as shown in Table 6, based on the constructed method, the failure probability (FP) up to t = 26.00 is obtained, as follows: Figure 7 As shown.
[0180] By further simulating future operating conditions through state transitions, the failure probability curve after time t is predicted. Using the obtained 100 failure probability prediction curves, the failure probability curve for any subsequent time t can be obtained. s The maximum and minimum failure probabilities of the tool at time t are used to obtain the lower limit failure probability curve and the upper limit failure probability curve, as shown below. Figure 7 As shown.
[0181] Table 6. Duration distribution of operating conditions
[0182]
[0183] Consider a future time t s From 100 predicted failure probability data points, we can obtain their failure probability histograms (FPHs), which can be used to predict the failure probability distribution at any future time. Figure 7 The FPH in the figure indicates that at time t s When the value is 27.40, the maximum possible failure probability lies between 0.51 and 0.52. Furthermore, this example evaluates the mode M0 of 100 predicted failure probability data points as the most likely failure probability curve for the predicted results, as shown below. Figure 8 As shown.
[0184]
[0185] in It is the group with the highest frequency in the histogram. and These are its lower and upper limits, respectively. d is the group distance under equal interval grouping, where... f m ,f m-1 ,f m+1 These are the frequencies of the most frequent group, the group preceding the most frequent group, and the group following the most frequent group, respectively. For any future time t... s By simulating, the mode M0 at all different times can be obtained, and the mode curve of the failure probability can be obtained, which represents the most likely failure probability curve under heterogeneous working conditions.
[0186] Finally, this example analyzes the Failure Time Histogram (FTH) under a given failure probability. It is assumed that when the failure probability reaches 0.999999, the machine tool will definitely fail, and the corresponding FTH is as follows: Figure 7 As shown. Based on Figure 7 As shown, FTH displays the distribution function of the tool's maximum remaining life. Based on Figure 7 According to FTH, the most likely failure time of the machine tool tool will fall between 33.4 and 33.5.
[0187] To address the shortcomings of current prediction methods in predicting tool failure probability and remaining life under heterogeneous operating conditions, this invention proposes a novel failure probability and remaining life prediction model. This method independently analyzes life data under homogeneous operating conditions, using it as prior knowledge to guide failure and remaining life prediction. Simultaneously, by constructing time-domain and amplitude-domain feature indices, a sensor data-based identification method is proposed to detect the similarity of sensor data at different times, thereby achieving effective identification of homogeneous operating conditions. Then, based on the proposed identification method, a failure prediction model based on machine tool condition identification under heterogeneous operating conditions is constructed. Furthermore, an online remaining life prediction method based on real-time historical operating conditions is established. Finally, application analysis is conducted based on real experimental data, demonstrating how the constructed method can complete tool failure probability diagnosis and remaining life prediction under heterogeneous operating conditions, and verifying the feasibility of the method.
[0188] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for fault diagnosis and remaining life prediction of machine tool cutting tools under heterogeneous working conditions, characterized in that, The steps are as follows: Step 1: Based on the homogeneous working condition classification method, conduct experiments on the machine operation process under different working conditions and collect the failure time data of machine tool tools under different working conditions; Step 2: Based on the collected machine tool failure time data, construct a one-dimensional homogeneous reliability model of the machine tool using the Weibull model; The method for constructing a one-dimensional homogeneous reliability model of machine tool cutting tools using the Weibull model is as follows: For the operation of a machine tool, all defined homogeneous working conditions constitute its possible set of working conditions, denoted as . ,in, This represents the total number of all possible homogeneous operating conditions in a machine tool. By performing independent model parameter estimation on the fault data under each homogeneous operating condition, a one-dimensional prediction model for machine tool faults under homogeneous operating conditions is obtained. ; The machine tool operates at any time under the same working conditions Subsequently, the failure probability and failure rate of the tool are denoted as follows: and ; Step 3: Identify heterogeneous operating conditions during machine tool operation based on real-time monitoring data from sensors; Step 4: Under heterogeneous working conditions, the one-dimensional homogeneous reliability model is mixed to establish a high-dimensional fault prediction model for machine tool tools under heterogeneous working conditions, and the real-time fault probability prediction is performed by combining the real-time heterogeneous working condition identification results. The high-dimensional fault prediction model for machine tool cutting tools under heterogeneous working conditions is as follows: ; in, For high-dimensional fault prediction models, The functional expression representing the high-dimensional fault prediction model. As of the current moment Historical operating condition sequence, As of the current moment The duration of the corresponding operating condition sequence; The method for real-time fault probability prediction includes: Using machine tool failure probability prediction under homogeneous operating conditions as prior knowledge, and based on the one-dimensional failure rate model under homogeneous operating conditions, a high-dimensional weighted failure rate model under heterogeneous operating conditions can be constructed: ; Where, τ i (t) represents the machine tool operating under the same conditions up to time t. pi The proportion of time spent working, and At any given time t, under any homogeneous operating condition c pi The cumulative time ratio τ i (t) is represented as: ; Among them, t c This is the starting point of the current homogeneous operating condition, a i This indicates the homogeneous operating condition c before the current homogeneous operating condition begins. pi The cumulative working hours, b i Indicates whether the current operating condition is a homogeneous operating condition (c). pi Judgment coefficient: b i =1 indicates that the current working condition is a homogeneous working condition c. pi Otherwise b i =0; According to the definition of failure rate, the following differential equation exists: ; Among them, f m (t) is the probability density function of machine tool failure under heterogeneous operating conditions, F m (t) is the cumulative distribution function of machine tool faults under heterogeneous operating conditions; considering F m (0)=0 and r m (0)=0, the actual cumulative distribution function of the machine tool under heterogeneous working conditions is: ; For any operating time interval [t1, t2] under the same operating conditions, there are two cases: (1) If b i If = 0, then the formula for calculating the actual cumulative distribution function can be obtained. ; in, This represents the shape parameters of a one-dimensional reliability model of machine tool cutting tools under the Weibull distribution. This represents the scale parameters of a one-dimensional reliability model for machine tools under the Weibull distribution; (2) If b i =1, and the formula for calculating the integral above is: ; Based on the above two scenarios, the integral formula for calculating the actual cumulative distribution function of the machine tool under heterogeneous working conditions can be summarized as follows: ; The integration range is [0, t1], [t1, t2], …, [t c [t] represents the continuous operating time of the machine tool under a certain homogeneous working condition, k=1,2,3,…K is the duration sequence, and K is the cumulative number of occurrences of all different homogeneous working conditions; furthermore, t0=0, t K-1 =t c , and t K =t; Step 5: Based on the real-time monitoring and updating of historical working condition information by sensors, the high-dimensional fault prediction model is updated according to the real-time updated historical working condition information. Based on the updated high-dimensional fault prediction model, the remaining life of machine tool tools is predicted.
2. The method for fault diagnosis and remaining life prediction of machine tool cutting tools under heterogeneous working conditions according to claim 1, characterized in that, The homogeneous operating condition classification method includes: In machine tools, the heterogeneity of operating conditions is represented by the following vector: ; in, for A 3D vector represents a set of working condition variables with discrete properties. The number of discrete attribute working condition variables; It is A 3D vector represents a set of operating condition variables with continuous attributes. It represents the number of continuous attribute operating condition variables; A set representing operating condition variables; The criteria for determining whether two operating conditions are identical are as follows: ; in, Indicates working conditions The values of discrete variables under the given conditions. Indicates working conditions The discrete variable takes values under the following conditions. For working conditions Normalized values of continuous attribute variables, For working conditions Normalized values of continuous attribute variables, and Both are vector distances measured in Euclidean geometry. To determine the threshold for homogeneous operating conditions; and These represent two operating conditions at different time periods.
3. The method for fault diagnosis and remaining life prediction of machine tool cutting tools under heterogeneous working conditions according to claim 1, characterized in that, In step three, the specific implementation method is as follows: Under given homogeneous working conditions The machine tool's reference sensor data is as follows: ; Where k is the number of sensors. For a given homogeneous working condition The data sample set obtained by the j-th sensor at different times, j=1,2,3,…,k Let the sensor data sample vector obtained by the j-th sensor at time l with a specific sampling frequency be represented as follows after normalizing its amplitude: ; During machine tool operation, based on monitoring data from k sensors, the real-time normalized monitoring data sample is obtained as follows: ; By comparing the similarity between sensor monitoring data D and reference sensor data D0, the current operating condition of the machine tool can be identified; the reference sensor sample Real-time monitoring samples of sensor j They are labeled as data 1 and 2 respectively; In the amplitude domain, the area enclosed by the probability density function curves of the reference sensor sample and the real-time monitoring sample is defined as its deviation D. Y This is used to estimate their similarity in the amplitude domain, and its calculation formula is: ; Where, f Y1 (y) is the probability density function of the normalized amplitude of the reference sensor data sample, f Y2 (y) is the probability density function of the normalized amplitude of the real-time monitoring sensor data sample, where y is the normalized amplitude; since D Y If ∈[0, 2], then the deviation of the normalization is: ; In the time domain, the cross-correlation coefficient can be used to measure the similarity between two time series, and its expression is: ; in, It is a reference sensor data sample The time series is denoted as time series 1; It is a real-time monitoring sensor data sample The time series with a delay of τ, denoted as time series 2, has its normalized cross-correlation coefficient expressed as: ; Among them, R 11 (0) is a reference sensor data sample. The autocorrelation coefficient, R 22 (0) is a sample of real-time sensor monitoring data. The autocorrelation coefficient; the maximum normalized cross-correlation coefficient obtained for different delay times τ. Represented as This can represent the best level of similarity between the two data samples; Based on the definitions of deviation and cross-correlation coefficient, the similarity between two sensor data samples is inversely proportional to the deviation and directly proportional to the cross-correlation coefficient. Therefore, the average difference between the cross-correlation coefficient and the deviation of k monitoring sensors is used to measure the similarity of sensor monitoring data under two operating conditions, i.e., to measure the similarity between the two operating conditions. The calculation formula is as follows: ; In addition, deviation If both the cross-correlation coefficient and the cross-correlation coefficient are normalized values, i.e., falling between [0, 1], then the two points ( , The Euclidean distance between (0, 1) and (0, 1) is used to measure the similarity of sensor monitoring data under two operating conditions, i.e., to measure the similarity between the two operating conditions. The calculation formula is: ; in, This indicates a similarity index calculated based on the difference. This represents a similarity index calculated based on Euclidean distance. It is the number of sensors used for similarity level assessment. It is a similarity evaluation standard based on the amplitude range. It is a time-domain based similarity evaluation standard; for a given similarity index SL, SL = SL dis Or SL=SL dif By comparing real-time monitored sensor data samples with a certain homogeneous operating condition c pi All reference sensor data samples D 0i By comparing the current real-time monitoring data sample with that of the same operating condition c, we can find the data. pi The level of most similarity is denoted as SL. i , indicating the current operating condition and the same operating condition c pi The similarity level between them; by comparing real-time monitoring data with all homogeneous operating conditions c pi By comparing the real-time monitoring data with all similar operating conditions, the similarity index value can be obtained. The following formula can then be used to determine which type of similar operating condition the processing stage containing the real-time monitoring data belongs to: ; in, This indicates that the current real-time monitoring data has been determined to be in the same type of working condition.
4. The method for fault diagnosis and remaining life prediction of machine tool cutting tools under heterogeneous working conditions according to claim 1, characterized in that, In step five, the specific implementation method is as follows: At any given time t, obtain real-time updated historical operating condition information, including operating condition sequence N. b (t) and its duration sequence T b (t); Based on the operating condition sequence, obtain the state transition matrix P between homogeneous operating conditions of the machine tool: ; Where, p ij The machine tool is in working condition c pi to c pj The transition probability; Based on the latest state transition matrix, the transition trajectory of future operating conditions is obtained through simulation. ; Based on real-time updated historical operating condition information, the time distribution characteristics of each homogeneous operating condition are updated in real time, and then the duration of different future operating conditions is simulated. ; Based on the Monte Carlo method, the possible life trajectory of machine tool tools is estimated, and the remaining tool life and its distribution function are predicted and analyzed by simulating the future working condition sequence of the machine tool.
