Service life prediction method, service life prediction device and service life prediction equipment for key components of machine

Abstract

The invention discloses a service life prediction method, a service life prediction device and service life prediction equipment for key components of a machine. The method for predicting the service life of the key component of the machine comprises the steps of obtaining working data of the key component of the machine; obtaining a statistical result according to the current data and the historical data; inputting the total used time length and the accumulated number of faults into a pre-trained linear regression model, outputting a first predicted number of faults of the key part in a first predicted time window, and obtaining a first predicted accumulated number of faults according to the first predicted number of faults; taking the predicted fault frequency and the predicted accumulated fault frequency corresponding to each predicted time window as a group of input features, and sequentially inputting the input features into a preset decision tree model for judgment according to the sequence of the predicted time windows; and when the judgment result output by the decision tree model is a replacement instruction, obtaining the number of completed loop iterations at the moment, and obtaining the remaining service life of the key part according to the number of loop iterations and the length of a single prediction time window.

Description

Technical Field

[0001] This application relates to the field of life prediction technology, and in particular to a life prediction method, life prediction device and life prediction equipment for key components of a machine tool. Background Technology

[0002] Computer Numerical Control (CNC) machine tools are core equipment in manufacturing. The performance degradation of key components (such as spindles, bearings, and transmission gears) directly affects machining accuracy and equipment reliability. In related technologies, these machines are equipped with Condition-Based Monitoring (CBM) systems. These systems can set fixed thresholds using sensors, triggering alarms when these limits are exceeded (e.g., an alarm when vibration acceleration exceeds a preset value), relying on manual judgment for subsequent actions. However, existing monitoring systems can only provide fault detection and cannot predict the failure of key components or their lifespan. This can easily lead to unplanned failures or damage to key components, affecting the stability and reliability of the machine tool's operation. Summary of the Invention

[0003] In order to overcome the above-mentioned technical defects and at least solve the technical problem that existing monitoring systems can only provide fault detection but cannot predict the faults and service life of key components, this application provides a method, device and equipment for predicting the service life of key components of a machine tool.

[0004] The method for predicting the lifespan of key components of a machine tool according to this application includes: acquiring operating data of the key components of the machine tool, the operating data including current data and historical data; obtaining statistical results based on the current data and the historical data, the statistical results including the total used time of the key components from the first time the key components were used to the end of the current time window and the cumulative number of failures within the total used time, wherein the current time window is a fixed-length and continuous time period; inputting the total used time and the cumulative number of failures into a pre-trained linear regression model to output a first predicted number of failures of the key components within the first prediction time window immediately following the current time window; adding the first predicted number of failures to the cumulative number of failures within the total used time to obtain a first predicted cumulative number of failures, wherein the output of the pre-trained linear regression model consists of at least an intercept, a first regression coefficient related to the total used time, and a second regression coefficient related to the cumulative number of failures within the total used time. The coefficients are jointly determined. When there are multiple fault types, the pre-trained linear regression model can be set separately for different fault types. Starting from the first prediction time window, multiple consecutive prediction time windows after the first prediction time window are iterated in a loop to obtain the number of predicted faults and the number of predicted cumulative faults corresponding to each prediction time window. The number of predicted faults and the number of predicted cumulative faults corresponding to each prediction time window are used as a set of input features and are sequentially input into the preset decision tree model for judgment according to the order of the prediction time windows. When there are multiple fault types, there are also multiple and corresponding numbers of predicted faults and the number of predicted cumulative faults. When the decision tree model outputs a replacement instruction, the number of times the linear regression model has been called during the iterative process of multiple consecutive prediction time windows is obtained. The number of times it has been called is used as the iteration count. Based on the iteration count and the length of a single prediction time window, the remaining service life of the key component is obtained.

[0005] In some implementations, acquiring the operational data of the key component includes: collecting current data and historical data of the key component; the current data includes: the current usage duration of the fault type of the key component from the start time of the current time window to the end time of the current time window, and the number of current faults occurring within the current usage duration; the historical data includes: the historical usage duration of the fault type of the key component from the first use of the key component to the start time of the current time window, and the number of historical faults occurring within the historical usage duration; compressing the current data to obtain current compressed data; buffering and transmitting the current compressed data using Apache Kafka, and using a clustering analysis algorithm to remove abnormal data in the current compressed data to obtain target data; and merging the target data and the historical data together as the operational data of the key component.

[0006] In some implementations, obtaining statistical results based on the current data and the historical data includes: if there are multiple fault types, obtaining all fault types occurring in the critical component within the current time window and the failure mode corresponding to each fault type based on the current data and the historical data; calculating the fault score for each failure mode corresponding to the fault type using the FMEA method; filtering out failure modes whose fault scores are not less than a preset score threshold, and designating these failure modes as critical failure modes; obtaining the number of faults of the fault type corresponding to the critical failure mode within the current time window of the critical component based on the working data, as the statistical fault count; and obtaining the statistical result based on the statistical fault count of the fault type corresponding to the critical failure mode and the historical fault count of the same fault type, wherein the statistical result includes the cumulative number of faults of the fault type corresponding to the critical failure mode within the current usage time.

[0007] In some implementations, the data types of the working data include temperature, rotational speed, vibration, and load; correspondingly, the fault types include temperature faults, rotational speed faults, vibration faults, and load faults; the historical fault counts include historical temperature fault counts, historical rotational speed fault counts, historical vibration fault counts, and historical load fault counts; the statistical fault counts include temperature fault counts, rotational speed fault counts, vibration fault counts, and load fault counts; and the cumulative fault counts include cumulative temperature fault counts, cumulative rotational speed fault counts, cumulative vibration fault counts, and cumulative load fault counts.

[0008] In some implementations, there are multiple fault types and multiple pre-trained linear regression models, each corresponding to one fault type. In each iteration of the cyclic iteration, for each fault type, the cumulative number of faults corresponding to that fault type and the current usage time are input into the corresponding linear regression model to obtain the predicted number of faults for that fault type within the prediction time window, and the predicted cumulative number of faults for that fault type is updated. The predicted number of faults and the predicted cumulative number of faults for each prediction time window are used as a set of input features and sequentially input into a preset decision tree model for judgment according to the order of the prediction time windows. When the decision tree model outputs a replacement instruction, the linear regression model obtained during the cyclic iteration of the multiple consecutive prediction time windows is used. The process involves determining the remaining service life of a critical component based on the number of times the regression model has been called, using this number as the iteration count, and then calculating the remaining service life of the critical component according to the iteration count and the length of a single prediction time window. This includes: using the predicted failure count and the predicted cumulative failure count for each failure type as a set of input features, and then sequentially inputting them into a preset decision tree model according to the order of the prediction time windows for judgment; when the decision tree model outputs a replacement instruction, obtaining the number of times the linear regression model has been called during the iterative process of multiple consecutive prediction time windows for that failure type, using this number as the iteration count, and then calculating the remaining service life of the critical component corresponding to that failure type according to the iteration count and the length of a single prediction time window. In some embodiments, the lifespan prediction method further includes: acquiring training parameters, the training parameters including the statistical number of failures of the key component within each historical time window, and the cumulative number of failures within the used time before any of the time windows, the statistical number of failures including the number of temperature failures, the number of speed failures, the number of vibration failures, and the number of load failures, the cumulative number of failures including the cumulative number of temperature failures, the cumulative number of speed failures, the cumulative number of vibration failures, and the cumulative number of load failures; and inputting any one of the statistical number of failures, any one of the cumulative number of failures, and the used time into a preset linear regression model for training, so as to obtain multiple pre-trained linear regression models.

[0009] In some implementations, the step of inputting any one of the statistical failure counts, any one of the cumulative failure counts, and the usage duration into a preset linear regression model for training to obtain multiple pre-trained linear regression models includes: processing any one of the statistical failure counts, any one of the cumulative failure counts, and the usage duration based on minimizing error to obtain an intercept and regression coefficients; and inputting the intercept and the regression coefficients into a preset linear regression model to obtain the pre-trained linear regression models.

[0010] In some embodiments, the lifespan prediction method further includes: the preset decision tree model is obtained by: acquiring multiple training features, the training features including the statistical number of failures of the key component in each historical time window, and the cumulative number of failures within the cumulative used time, the statistical number of failures including the number of temperature failures, speed failures, vibration failures, and load failures, and the cumulative number of failures including the cumulative number of temperature failures, speed failures, vibration failures, and load failures; calculating the information gain of the multiple training features using a preset algorithm; and selecting the training feature with the largest information gain. The training features of each split node are used as split nodes; and the training features of each split node are recursively split, and the process returns to continue executing the step of calculating the information gain of multiple training features using a preset algorithm, until the recursive split stops after the recursive termination condition is met, so as to obtain a preset decision tree model; wherein, the recursive termination condition is: when the depth of the recursive split reaches the maximum depth threshold determined by cross-validation or the number of training samples contained in the current split node is lower than the preset minimum number of samples threshold, the split stops and the current split node is marked as a leaf node, and the output of the leaf node is "replace instruction" or "do not replace instruction", thereby obtaining the preset decision tree model.

[0011] The life prediction device for key components of a machine tool according to this application includes a data acquisition layer, a processing layer, a first decision layer, and a second decision layer. The data acquisition layer is configured to acquire operating data of the key components of the machine tool, including current data and historical data. The processing layer is configured to acquire statistical results based on the current data and historical data. The statistical results include the total used time of the key component from the first time the key component was used until the end of the current time window, and the cumulative number of failures within the total used time. The current time window is a fixed-length and continuous time period. The first decision layer is configured to input the total used time and the cumulative number of failures into a pre-trained linear regression model to output a first predicted number of failures of the key component within the first prediction time window immediately following the current time window. The first predicted number of failures is added to the cumulative number of failures within the total used time to obtain a first predicted cumulative number of failures. The output of the pre-trained linear regression model consists at least of an intercept, a first regression coefficient related to the total used time, and a second decision layer. The second regression coefficient, related to the cumulative number of failures within the total used time, is jointly determined. When there are multiple types of failures, the pre-trained linear regression model can be set separately for different failure types. The second decision layer is configured to iterate through multiple consecutive prediction time windows starting from the first prediction time window to obtain the predicted number of failures and the predicted cumulative number of failures for each prediction time window. The predicted number of failures and the predicted cumulative number of failures for each prediction time window are used as a set of input features and input into a preset decision tree model in the order of the prediction time windows for judgment. When the decision tree model outputs a replacement instruction, the number of times the linear regression model has been called during the iterative process of the multiple consecutive prediction time windows is obtained. The number of times it has been called is used as the iteration count, and the remaining service life of the key component is obtained based on the iteration count and the length of a single prediction time window.

[0012] The life prediction device for key machine components according to embodiments of this application includes a memory and a processor, wherein the memory stores instructions. The instructions stored in the memory are executed by the processor to implement the life prediction method as described in any of the above embodiments.

[0013] In the life prediction method, life prediction device, and life prediction equipment for key machine components according to the embodiments of this application, statistical results are obtained based on the current and historical data of the key components of each machine. The statistical results include the total used time from the first use of the key component to the end of the current time window, and the cumulative number of failures within the total used time. A pre-trained linear regression model is used to obtain the first predicted number of failures for the key component within the first prediction time window immediately following the current time window. This first predicted number of failures is added to the cumulative number of failures within the total used time to obtain the first predicted cumulative number of failures. Starting from the first prediction time window, multiple consecutive prediction time windows following the first prediction time window are iteratively applied to obtain... The system obtains the predicted number of failures and the predicted cumulative number of failures for each prediction time window. These are then used as a set of input features, sequentially input into a pre-defined decision tree model according to the order of the prediction time windows. When the decision tree model outputs a replacement instruction, the system obtains the number of times the linear regression model has been called during the iterative process of multiple consecutive prediction time windows. This number of calls is used as the iteration count. Finally, based on the iteration count and the length of a single prediction time window, the remaining service life of the critical components is determined. This provides data support for failure prevention, enables predictive maintenance, avoids over-maintenance and sudden downtime, reduces maintenance costs, and improves the stability and reliability of machine operation.

[0014] Additional aspects and advantages of embodiments of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of embodiments of this application. Attached Figure Description

[0015] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, wherein: Figure 1 This is a flowchart illustrating a method for predicting the lifespan of key machine components according to certain embodiments of this application. Figure 2 This is a schematic diagram of the life prediction device for a key component of a machine tool according to certain embodiments of this application. Figure 3 This is a structural schematic diagram of a life prediction device for a key component of a machine tool according to certain embodiments of this application; Figure 4 This is a flowchart illustrating a method for predicting the lifespan of key machine components according to certain embodiments of this application. Figure 5This is a flowchart illustrating a method for predicting the lifespan of key machine components according to certain embodiments of this application. Figure 6 This is a flowchart illustrating a method for predicting the lifespan of key machine components according to certain embodiments of this application. Figure 7 This is a flowchart illustrating a method for predicting the lifespan of key machine components according to certain embodiments of this application. Figure 8 This is a flowchart illustrating a method for predicting the lifespan of key machine components according to certain embodiments of this application. Figure 9 This is a flowchart illustrating a method for predicting the lifespan of key machine components according to certain embodiments of this application.

[0016] Explanation of key component symbols: 100 lifetime prediction devices, 110 memory, 130 processors; 10 Lifetime prediction device, 11 Acquisition layer, 13 Processing layer, 15 First decision layer, 17 Second decision layer. Detailed Implementation

[0017] The embodiments of this application will be further described below with reference to the accompanying drawings. The same or similar reference numerals in the drawings denote the same or similar elements or elements having the same or similar functions throughout. Furthermore, the embodiments of this application described below with reference to the accompanying drawings are exemplary and are only used to explain the embodiments of this application, and should not be construed as limiting this application.

[0018] Computer Numerical Control (CNC) machine tools are core equipment in manufacturing. The performance degradation of key components (such as spindles, bearings, and transmission gears) directly affects machining accuracy and equipment reliability. In related technologies, machine tools are equipped with condition-based monitoring (CBM) systems. These systems can set fixed thresholds using sensors, triggering alarms when these thresholds are exceeded (e.g., an alarm when vibration acceleration exceeds a preset value), relying on manual judgment for subsequent actions. However, existing monitoring systems can only provide fault detection and cannot predict the failure and lifespan of key components, easily leading to unplanned failures or damage, affecting the stability and reliability of machine tool operation. To address this problem, this application provides a method for predicting the lifespan of key machine tool components (…). Figure 1 As shown), life prediction device 10 for key components of the machine tool. Figure 2 As shown), life prediction equipment 100 for key components of the machine tool. Figure 3 (as shown), computer-readable storage media and computer program products.

[0019] The life prediction method for key machine tool components according to the embodiments of this application includes: 01: Obtain the operating data of key components of the machine tool, including current data and historical data; 02: Obtain statistical results based on current and historical data. The statistical results include the types of failures that occurred in key components, the total usage time from the time the key component was first used to the end of the current time window, and the cumulative number of failures within the total usage time. The current time window is a fixed-length and continuous time period. 03: Input the total used time and cumulative failure count into a pre-trained linear regression model to output the first predicted failure count of the key component within the first predicted time window immediately following the current time window. Add the first predicted failure count to the cumulative failure count within the total used time to obtain the first predicted cumulative failure count. The output of the pre-trained linear regression model is determined by at least the intercept, the first regression coefficient related to the total used time, and the second regression coefficient related to the cumulative failure count within the total used time. When there are multiple failure types, the pre-trained linear regression model can be set separately for different failure types; and 04: Starting from the first prediction time window, iterate through multiple consecutive prediction time windows after the first prediction time window to obtain the predicted number of faults and the predicted cumulative number of faults for each prediction time window; use the predicted number of faults and the predicted cumulative number of faults for each prediction time window as a set of input features, and input them sequentially into a preset decision tree model for judgment according to the order of the prediction time windows; wherein, when there are multiple types of faults, there are also multiple predicted number of faults and the predicted cumulative number of faults, which correspond one-to-one; in this application, the predicted number of faults is the number of newly added faults; When the decision tree model outputs a replacement instruction, the number of times the linear regression model has been called during the iterative process of the multiple consecutive prediction time windows is obtained. The number of times the model has been called is taken as the iteration count. Based on the iteration count and the length of a single prediction time window, the remaining service life of the key component is obtained.

[0020] Please combine Figure 2The aforementioned method for predicting the lifespan of key machine components can be applied to a lifespan prediction device 10 for key machine components. The lifespan prediction device 10 for key machine components according to this application includes a data acquisition layer 11, a processing layer 13, a first decision layer 15, and a second decision layer 17. The data acquisition layer 11 is configured to acquire operating data of the key machine components, including current data and historical data. The processing layer 13 is configured to acquire statistical results based on the current data and historical data. The statistical results include the types of failures occurring in the key components, the total used time from the first use of the key components to the end of the current time window, and the cumulative number of failures within the total used time. The current time window is a fixed-length and continuous time period. The first decision layer 15 is configured to input the total used time and cumulative failure count into a pre-trained linear regression model to output the first predicted failure count of the key component within the first prediction time window immediately following the current time window. The first predicted failure count is then added to the cumulative failure count within the total used time to obtain the first predicted cumulative failure count. The output of the pre-trained linear regression model is determined at least by the intercept, the first regression coefficient related to the total used time, and the second regression coefficient related to the cumulative failure count within the total used time. When there are multiple failure types, the pre-trained linear regression model can be set separately for different failure types. The second decision layer 17 is configured to iterate through multiple consecutive prediction time windows following the first prediction time window. The process involves iterative looping (i.e., using the state at the end of the current time window as the initial condition, and sequentially iterating over multiple consecutive prediction time windows, including the first prediction time window), to obtain the predicted number of failures and the predicted cumulative number of failures for each prediction time window. These are then used as a set of input features, sequentially input into a pre-defined decision tree model according to the order of the prediction time windows. When the decision tree model outputs a replacement instruction, the number of times the linear regression model has been called during the iterative looping of the multiple consecutive prediction time windows is obtained. This number of calls is used as the iteration count, and the remaining service life of the critical component is determined based on the iteration count and the length of a single prediction time window.

[0021] In this invention, a time window refers to a preset, fixed-length continuous time period used to segment and model the operating status of the key components. The historical time window represents a time window that has ended before the life prediction is initiated and is used for model training. The current time window represents a time window that has just ended when the life prediction is initiated and is used to provide the initial state for prediction. The prediction time window represents future time windows that are simulated sequentially after the life prediction begins, and the number of failures is recursively predicted by a linear regression model. The three time windows have the same preset length and are arranged continuously on the time axis without overlapping each other.

[0022] Please combine Figure 3 The aforementioned method for predicting the lifespan of critical machine components can be applied to a lifespan prediction device 100 for critical machine components. One embodiment of the lifespan prediction device 100 for critical machine components includes a memory 110 and a processor 130. The memory 110 stores instructions. The instructions stored in the memory 110 are executed by the processor 130 to implement the lifespan prediction method in steps 01, 02, 03, and 04. That is, the processor 130 is used to: acquire working data of key components of the machine, including current data and historical data; acquire statistical results based on the current data and historical data, the statistical results including the total usage time of the key component from the time the key component was first used to the end of the current time window and the cumulative number of failures within the total usage time, wherein the current time window is a fixed-length and continuous time period; input the total usage time and the cumulative number of failures into a pre-trained linear regression model, output the first predicted number of failures of the key component within the first predicted time window immediately following the current time window, add the first predicted number of failures to the cumulative number of failures within the total usage time to obtain the first predicted cumulative number of failures, the output of the pre-trained linear regression model consisting of at least the intercept and the sum of the values ​​relative to the total usage time. The first regression coefficient and the second regression coefficient, which are related to the cumulative number of failures within the total used time, jointly determine the remaining service life of the critical component. When there are multiple types of failures, the pre-trained linear regression model can be set separately for different failure types. Starting from the first prediction time window, multiple consecutive prediction time windows after the first prediction time window are iterated to obtain the predicted number of failures and the predicted cumulative number of failures for each prediction time window. The predicted number of failures and the predicted cumulative number of failures for each prediction time window are used as a set of input features and are input into the preset decision tree model in the order of the prediction time windows for judgment. When the decision tree model outputs a replacement instruction, the number of iterations completed at this time is obtained, and the remaining service life of the critical component is obtained based on the number of iterations and the length of a single prediction time window.

[0023] Specifically, a Computer Numerical Control (CNC) machine tool is a machining tool that can be controlled by pre-programmed computer software. Users can control the machining parameters and paths of the cutting tools through pre-edited programs, thereby performing precise cutting, carving, drilling, and other machining processes on substrates (such as metals, plastics, and wood). CNC machine tools (hereinafter referred to as "machine tools") have many advantages, including high precision, good repeatability, high machining efficiency, high automation, and high flexibility, and are widely used in various machining processes. However, during prolonged use, critical components of the machine tool (such as spindles, bearings, and transmission gears) may malfunction, such as spindle abnormalities or bearing abnormalities, leading to a decrease in the yield rate of machining. This application provides a method for predicting the lifespan of critical components of a machine tool. This method can predict the remaining lifespan of critical components based on their operating data, providing data support for fault prevention and avoiding sudden downtime. For ease of understanding, the following embodiments use a spindle as an example of a critical component.

[0024] The life prediction device 100 is a device used to predict the lifespan of critical components of a machine tool. The memory 110 stores instructions corresponding to the life prediction, and the processor 130 executes the instructions stored in the memory 110 to enable the life prediction device 100 to implement the life prediction method. The memory 110 and the processor 130 are electrically connected and connected to various parts of the entire life prediction device 100 via various interfaces and lines. The processor 130 executes methods O1, O2, O3, and O4 by running or loading instructions stored in the memory 110 and calling data stored in the memory 110, thereby realizing the lifespan prediction of critical components.

[0025] During the processing of substrates by the machine tool, key components generate certain operational data. This operational data includes, but is not limited to, the temperature, rotational speed, vibration, and load of the key components. The operational data of key components reflects their state (such as operational status and health status). The lifespan prediction method of this application can acquire the operational data of key components for each machine tool and use this data to predict the remaining service life of the key components, thus helping to improve the accuracy of lifespan prediction.

[0026] In step 02, the total used time can be the cumulative used time of the machine's critical components from the start of use to the end of the current time window, and the cumulative number of failures can be the total number of failure types of the critical components occurring within the total used time. The processor 130 can obtain the total used time and cumulative number of failures based on current and historical data to facilitate subsequent lifespan prediction. It should be noted that in some embodiments, failure types include, but are not limited to, abnormal spindle (critical component) temperature, unstable spindle speed, excessive spindle load, and excessive spindle vibration. The current time window can be a known, fixed-length, continuous time period. For example, the length of the current time window can be one day or one week.

[0027] In step 03, the processor 130 can input the total used time and cumulative number of failures obtained in step 02 into a pre-trained linear regression model for prediction, so as to obtain the first predicted number of failures in the first prediction time window immediately following the current time window, and add the first predicted number of failures to the cumulative number of failures in the total used time to obtain the first predicted cumulative number of failures.

[0028] Optionally, a linear regression model is a statistical model used to model and analyze linear relationships between variables. It assumes a linear relationship between the dependent variable (target variable) and one or more independent variables (feature variables) and predicts data by fitting a straight line (or hyperplane). That is, a linear regression model can predict the number of failures in the next time window based on two independent variables: total usage time and cumulative failure count. A linear regression model can be a fixed model set at the factory or a variable model that operators can adjust and optimize.

[0029] Alternatively, the pre-trained linear regression model can be: ; in, For the intercept term, X1 represents the total usage time, and X2 represents the cumulative number of failures. The regression coefficient of the independent variable X1. The regression coefficients are for the independent variable X2. It should be noted that in a pre-trained linear regression model, the intercept (...) ), and the first regression coefficient related to the total used time ( ), and the second regression coefficient related to the cumulative number of failures over the total usage time ( All of these are known values.

[0030] Specifically, when the total used time is used as the independent variable X1 and the cumulative number of failures is used as the independent variable X2, the linear regression model can output the first predicted number of failures Y in the first predicted time window immediately following the current time window, and the processor 130 can add the first predicted number of failures Y to the cumulative number of failures in the total used time to obtain the first predicted cumulative number of failures.

[0031] Optionally, when there are multiple fault types, there can also be multiple pre-trained linear regression models. By inputting the cumulative number of faults corresponding to different fault types and the total used time into the corresponding linear regression models, the first predicted number of faults corresponding to different fault types can be obtained. The processor 130 can add the first predicted number of faults corresponding to different fault types to the cumulative number of faults of the corresponding fault types within the total used time to obtain the first predicted cumulative number of faults.

[0032] In step 04, after iterating through multiple consecutive prediction time windows, the processor 130 obtains the predicted number of failures and the predicted cumulative number of failures for each prediction time window. The processor 130 then uses these figures as a set of input features and sequentially inputs them into a preset decision tree model for judgment, according to the order of the prediction time windows. When the decision tree model outputs a replacement instruction, the processor obtains the number of times the linear regression model has been called during the iterative process of the multiple consecutive prediction time windows. This number is used as the iteration count, and the remaining service life of the critical component is obtained based on the iteration count and the length of a single prediction time window. For example, if the prediction time window is 1 day long, and the linear regression model iterates 5 times when the decision tree model outputs a replacement instruction, the processor 130 can determine that the remaining service life of the critical component is 5 days. This method automatically generates the remaining service life of the critical component by predicting the number of failures through linear regression and recursively determining the replacement timing using a decision tree, avoiding decision bias caused by human experience and reducing maintenance costs.

[0033] It should be noted that the decision tree model is a supervised learning model that uses a tree structure (nodes and branches) to classify or regress data. It can progressively divide data into different subsets through a series of rules and ultimately achieve the prediction target. The decision tree model can be a fixed model set at the factory or a variable model that can be adjusted and optimized by the operator.

[0034] For example, the current time window can be a fixed-length (1 day) and continuous time period. In this case, when the total used time and the cumulative number of failures are input into the pre-trained linear regression model for prediction, the linear regression model can predict the first predicted number of failures in the first prediction time window (the next day), and the processor 130 can obtain the first predicted cumulative number of failures by adding the first predicted number of failures to the cumulative number of failures. The processor 130 inputs the first predicted number of failures and the first predicted cumulative number of failures into the decision tree model. If the decision tree model outputs a decision that the instruction should not be replaced, the linear regression model can predict the second predicted number of failures for the next prediction time window (the next day) based on the first predicted cumulative number of failures and the total service life. The processor 130 can also obtain the second predicted cumulative number of failures by adding the second predicted number of failures and the cumulative number of failures. The processor 130 inputs the second predicted number of failures and the second predicted cumulative number of failures into the decision tree model and repeats the above steps until the decision tree model outputs a decision that the instruction should be replaced. At this time, the processor obtains the number of times the linear regression model has been called during the iterative process of multiple consecutive prediction time windows. The number of times the model has been called is used as the iteration count. Based on the iteration count and the length of a single prediction time window, the remaining service life of the critical component is obtained.

[0035] In the method for predicting the lifespan of key machine components according to the embodiments of this application, statistical results are obtained based on the current and historical data of the key components of each machine. The statistical results include the types of failures that occur in the key components, the total used time from the first use of the key components to the end of the current time window, and the cumulative number of failures within the total used time. A pre-trained linear regression model is then used to obtain the first predicted number of failures for the key components within the first prediction time window immediately following the current time window. The first predicted number of failures is added to the cumulative number of failures within the total used time to obtain the first predicted cumulative number of failures. Starting from the first prediction time window, multiple consecutive failures after the first prediction time window are then analyzed. The prediction time window is iterated repeatedly to obtain the predicted number of failures and the predicted cumulative number of failures for each prediction time window. The predicted number of failures and the predicted cumulative number of failures for each prediction time window are used as a set of input features and are sequentially input into a preset decision tree model for judgment according to the order of the prediction time windows. When the decision tree model outputs a replacement instruction, the number of iterations completed at this time is obtained. Finally, based on the number of iterations and the length of a single prediction time window, the remaining service life of the key components is obtained. This can provide data support for failure prevention, realize predictive maintenance, avoid over-maintenance and sudden failure downtime, reduce maintenance costs, and improve the stability and reliability of machine operation.

[0036] In some implementations, 01: Acquiring operational data of key components, including: 011: Collect current and historical data of key components. Current data includes: the current usage duration of the key component's failure type from the start time of the current time window to the end time of the current time window, and the number of current failures within the current usage duration; historical data includes: the historical usage duration of the key component's failure type from the time the key component was first used to the start time of the current time window, and the number of historical failures within the historical usage duration. 012: Compress the current data to obtain the current compressed data; 013: Apache Kafka is used to buffer and transmit the current compressed data, and a clustering analysis algorithm is used to remove outliers from the compressed data to obtain the target data; and 014: Merge target data and historical data to serve as working data for key components.

[0037] Please combine Figure 2 The acquisition layer 11 is also configured to: acquire current and historical data of key components. The current data includes: the current usage duration of the key component's failure type from the start time of the current time window to the end time of the current time window, and the number of current failures occurring within the current usage duration; the historical data includes: the historical usage duration of the key component's failure type from the first time the key component was used to the start time of the current time window, and the number of historical failures occurring within the historical usage duration; compress the current data to obtain current compressed data; buffer and transmit the current compressed data using Apache Kafka, and use clustering analysis algorithms to remove abnormal data in the current compressed data to obtain target data; and merge the target data and historical data together as the working data of the key component.

[0038] Please combine Figure 3The processor 130 in the life prediction device 100 of this application embodiment is also used to execute the life prediction methods in 011, 012, 013 and 014. That is, the processor 130 is also used to: collect current data and historical data of key components. The current data includes: the current usage duration of the fault type of the key component from the start time of the current time window to the end time of the current time window, and the number of current faults that occurred within the current usage duration; the historical data includes: the historical usage duration of the fault type of the key component from the first time the key component was used to the start time of the current time window, and the number of historical faults that occurred within the historical usage duration; compress the current data to obtain current compressed data; buffer and transmit the current compressed data using Apache Kafka, and remove abnormal data in the current compressed data using a clustering analysis algorithm to obtain target data; and merge the target data and historical data together as the working data of the key component.

[0039] Specifically, it should be noted that since the operating data of key components comes from multiple different sensors, and the sampling frequencies and protocols of these sensors differ significantly (such as Modbus / TCP, OPC UA), data timing misalignment is likely to occur when the processor 130 simultaneously collects multiple types of data (vibration, temperature, rotational speed, load, etc.). Therefore, directly using the raw operating data for processing may affect the accuracy of subsequent lifespan prediction. The processor 130 needs to process the collected raw data, obtain operating data based on the processed data, and then use the operating data for subsequent analysis.

[0040] At this time, the processor 130 can collect the current data and historical data of the key components, compress the current data to obtain the current compressed data, use Apache Kafka to buffer and temporarily store the current compressed data, and use cluster analysis algorithm to remove abnormal data in the current compressed data to obtain the target data. Finally, the target data and historical data are merged to serve as the working data of the key components.

[0041] Therefore, in the lifetime prediction method of this application, the processor 130 can use the Apache Kafka distributed queue to buffer and temporarily store high-frequency data, and combine it with a clustering analysis algorithm to remove abnormal data in the current compressed data. This ensures the real-time acquisition of multi-source heterogeneous data, eliminates the impact of sampling frequency and protocol differences on the data, and removes false alarm data induced by sensor failure or environmental factors, improving the accuracy of working data and thus improving the accuracy and reliability of subsequent lifetime prediction. It should be noted that in some embodiments, the clustering analysis algorithm can be DBSCAN density clustering (ε=0.5, MinPts=5).

[0042] Optionally, when removing abnormal data, the maintenance records corresponding to each fault can be combined, and the fault causes analyzed and recorded by the operators can be used to clearly identify faults caused by non-critical components (spindle) directly, such as sensor faults or environmental abnormalities.

[0043] Optionally, in addition to the above data, historical data may also include the time of failure, duration of failure, fault code, work order data (work order number, maintenance personnel, replacement time, disassembly time, spare part number, maintenance measures, and replacement part model).

[0044] In some implementations, 02: Obtain statistical results based on current and historical data, including: 021: There are multiple types of faults. Based on current and historical data, obtain all fault types of critical components that occur within the current time window and the corresponding failure modes for each fault type. 022: Using the FMEA method, calculate the failure score for each failure mode corresponding to each failure type; 023: Filter all failure modes with a fault score not less than a preset score threshold, and designate the failure modes with a fault score not less than the preset score threshold as critical failure modes. 024: Based on the working data, obtain the number of failures of the corresponding failure type of the critical failure mode in the current time window of the critical component, and use this as the statistical failure count; and 025: Obtain statistical results based on the number of failures corresponding to the critical failure mode and the historical number of failures of the same failure type. The statistical results include the cumulative number of failures of the critical failure mode corresponding to the failure type within the current usage time.

[0045] Please combine Figure 2 The processing layer 13 is also configured to: handle multiple fault types; based on current and historical data, obtain all fault types occurring in the critical component within the current time window and the corresponding failure modes for each fault type; calculate the fault score for each failure mode corresponding to the failure mode using the FMEA method; filter all failure modes whose fault scores are not less than a preset score threshold, and designate the failure modes with fault scores not less than the preset score threshold as critical failure modes; obtain the number of faults corresponding to each critical failure mode within the current time window of the critical component based on the working data, as the statistical fault count; and obtain statistical results based on the statistical fault count of each critical failure mode and the historical fault count of the same fault type, including the cumulative fault count of each critical failure mode within the current used time.

[0046] Please combine Figure 5The processor 130 in the life prediction device 100 of this application embodiment is also used to execute the life prediction methods in 021, 022, 023, 024 and 025. That is, the processor 130 is also used to: obtain all failure types of key components within the current time window and the failure mode corresponding to each failure type based on current data and historical data; calculate the failure score of the failure mode corresponding to each failure type using the FMEA method; filter failure modes with failure scores not less than a preset score threshold among all failure modes, and take the failure modes with failure scores not less than the preset score threshold as key failure modes; obtain the number of failures of the failure type corresponding to each key failure mode within the current time window of the key component based on working data, as the statistical failure count; and obtain statistical results based on the statistical failure count of the failure type corresponding to each key failure mode and the historical failure count of the same failure type. The statistical results include the cumulative number of failures of the failure type corresponding to each key failure mode within the current used time.

[0047] Specifically, it should be noted that since different failure types have different impacts on the service life of critical components, providing all failure type data to the linear regression model and decision tree model to predict the remaining service life of critical components would result in an excessively large amount of data and low prediction accuracy. Therefore, the processor 130 needs to evaluate and classify failure types to extract critical failure types as the data source for subsequent service life prediction.

[0048] At this point, in step 021, the processor 130 can analyze the current data based on pre-stored information (such as the spindle design and maintenance manual) to obtain the failure modes corresponding to all faults occurring in key components within the current time window. It should be noted that, based on the equipment manufacturer's maintenance manual and professional technical guidance, a set of fault types closely related to spindle life is defined. The spindle design and maintenance manual lists various common fault causes that significantly affect spindle life. These causes include, but are not limited to, bearing wear, seal failure, thermal deformation, unbalanced vibration, and material fatigue fracture. When these faults occur, the machine tool will respond immediately, triggering a series of fault prompts, paying particular attention to key fault types such as abnormal spindle temperature, unstable spindle speed, excessive spindle load, and excessive spindle vibration.

[0049] In step 022, the processor 130 uses FMEA (Failure Mode and Effects Analysis) to evaluate all failure modes and obtain failure scores for each. Specifically, for each failure mode of each failure type, its potential impact (such as performance degradation, increased downtime, increased maintenance costs, etc.), frequency of occurrence, and ease of detection are listed. By comprehensively evaluating the RPN (Risk Priority Number, usually determined by the product of severity (S), frequency (O), and detectability (D), these failure types are ranked according to priority. It should be noted that FMEA is a commonly used risk assessment method whose purpose is to identify potential failure modes and assess their impact on system performance; in FMEA, the Risk Priority Number (RPN) is an indicator used to assess and rank the risk severity of failure modes.

[0050] In steps 023 and 024, the processor 130 can filter all failure modes whose failure scores are not less than (greater than or equal to) a preset score threshold. The processor 130 identifies these failure modes as critical failure modes and obtains the statistical number of failures for each critical failure mode within the current time window based on the operating data. In other words, the processor 130 can accurately identify and extract the failure types most closely related to the lifespan prediction of critical components and obtain the statistical number of failures for each critical failure type within the current time window. Optionally, the preset score threshold can be known data; that is, the preset score threshold can be data set before the processor 130 leaves the factory, or data input manually after the processor 130 leaves the factory.

[0051] In step 025, the processor 130 can obtain statistical results based on the statistical number of failures corresponding to each critical failure mode and the historical number of failures of the same failure type. That is, the statistical results are obtained based on the statistical number of failures corresponding to the critical failure type and the historical number of failures corresponding to the critical failure type. The statistical results include the cumulative number of failures of the critical components of each machine for the fault type within the current usage time.

[0052] Thus, in the lifetime prediction method of this application, the processor 130 can evaluate and classify fault types to extract key fault types (fault types corresponding to key failure modes) as the data source for subsequent lifetime prediction. This not only improves the accuracy of lifetime prediction but also reduces the amount of data in subsequent processing, lowers the data processing difficulty, and helps improve prediction speed. Furthermore, based on the FMEA framework, different key components such as spindles, bearings, and gearboxes can be quickly adapted by adjusting feature weights without redeveloping the model architecture, flexibly adapting to different key components and improving scalability.

[0053] Furthermore, the processor 130 obtains the statistical number of failures corresponding to the critical failure mode within the current time window based on the current data. Therefore, the processor 130 can obtain the statistical number of failures within the current time window in real time. For example, the processor 130 can obtain the daily statistical number of failures in real time, thereby making the prediction of the lifespan of subsequent critical components more accurate and timely.

[0054] In some implementations, the data types of the working data include temperature, rotational speed, vibration, and load; correspondingly, the fault types include temperature faults, rotational speed faults, vibration faults, and load faults; the historical fault counts include the historical temperature fault counts, historical rotational speed fault counts, historical vibration fault counts, and historical load fault counts; the statistical fault counts include the temperature fault counts, rotational speed fault counts, vibration fault counts, and load fault counts; and the cumulative fault counts include the cumulative temperature fault counts, cumulative rotational speed fault counts, cumulative vibration fault counts, and cumulative load fault counts.

[0055] Specifically, in some embodiments, the processor 130 can synchronously acquire operating data of key components through multiple sensors. For example, the processor 130 can acquire vibration data of key components through vibration sensors (such as triaxial accelerometers), acquire temperature data of key components through temperature sensors, acquire rotational speed data of key components through speed sensors (such as photoelectric encoders), and acquire load data of key components through load sensors (such as strain gauge force sensors).

[0056] Optionally, the acquisition frequencies of multiple sensors can be the same or different. For example, the acquisition frequency for vibration data can be 200Hz, while the acquisition frequency for temperature data, load data, and rotational speed data can be 50Hz. The data format can be: time-series data stored as a Parquet file (e.g., compression ratio 4:1), including timestamps (accurate to milliseconds), sensor ID, and raw signal values. Optionally, the life prediction device 100 can be configured with interfaces; for example, the interfaces include 4 USB 3.0 ports and 2 Gigabit Ethernet ports, supporting Modbus / TCP and OPCUA protocols.

[0057] In some implementations, there are multiple fault types and multiple pre-trained linear regression models, each corresponding to a fault type. In this application, in each iteration of the loop, the cumulative number of predicted faults in the previous prediction time window is used as the current input to predict the number of faults in subsequent time windows in a recursive manner. For each fault type, the cumulative number of faults corresponding to the fault type and the current used duration are input into the corresponding linear regression model to obtain the predicted number of faults for that fault type in the prediction time window, and the predicted cumulative number of faults for that fault type is updated. 04: The predicted number of failures and the predicted cumulative number of failures for each prediction time window are used as a set of input features. These are then sequentially input into a pre-defined decision tree model for judgment, according to the order of the prediction time windows. When the decision tree model outputs a replacement instruction, the number of completed iterations is obtained, and the remaining service life of the critical component is calculated based on the number of iterations and the length of a single prediction time window. This includes: 041: The predicted number of failures for each type of failure and the predicted cumulative number of failures for each type of failure are taken as a set of input features and then input into the preset decision tree model in the order of the prediction time window for judgment. 042: When the decision tree model outputs a replacement instruction, obtain the number of iterations completed for that fault type, and based on the number of iterations and the length of a single prediction time window, obtain the remaining service life of the corresponding key component.

[0058] Please combine Figure 2 The second decision layer 17 is also configured to: take the predicted number of failures for each type of failure and the predicted cumulative number of failures for each type of failure as a set of input features, and input them into the preset decision tree model in the order of the prediction time window for judgment; when the decision tree model outputs a replacement instruction, obtain the number of loop iterations completed for that type of failure, and obtain the remaining service life of the corresponding key component based on the number of loop iterations and the length of a single prediction time window.

[0059] Please combine Figure 3In the life prediction device 100 of this application embodiment, the processor 130 is also used to execute the life prediction methods in 041 and 042. That is, the processor 130 is also used to: take the predicted number of failures and the predicted cumulative number of failures for each type of failure as a set of input features, and input them into a preset decision tree model for judgment according to the order of the prediction time window; when the judgment result output by the decision tree model is a replacement instruction, obtain the number of loop iterations that the failure type has completed at this time, and obtain the remaining service life of the key component corresponding to the failure type according to the number of loop iterations and the length of a single prediction time window.

[0060] Specifically, based on the above, the fault types include temperature faults, speed faults, vibration faults, and load faults. Therefore, there are four pre-trained linear regression models, each corresponding to one of the four fault types: temperature fault, speed fault, vibration fault, and load fault. In each iteration of the cyclical iteration, each linear regression model obtains the predicted fault count for the corresponding fault type within the prediction time window. The processor 130 updates the predicted cumulative fault count for the corresponding fault type based on the predicted fault count. Thus, within each prediction time window, the four linear regression models obtain four predicted fault counts and four predicted cumulative fault counts for the four fault types.

[0061] The processor 130 can take the four predicted failure counts and four predicted cumulative failure counts corresponding to each prediction time window as a set of input features, and input them into the decision tree model for judgment according to the order of the prediction time windows. When the decision tree model outputs a replacement instruction, it obtains the number of iterations completed by any linear regression model at this time, and obtains the remaining service life of the critical component based on the number of iterations and the length of a single prediction time window. In this way, by predicting the number of failures through linear regression and recursively determining the replacement timing through the decision tree, the remaining service life of the critical component is automatically generated, avoiding decision-making bias caused by human experience and reducing maintenance costs.

[0062] In some implementations, the lifetime prediction method further includes: 051: Obtain training parameters, including the statistical number of failures of key components within each historical time window, and the cumulative number of failures within the usage time prior to any given time window. The statistical number of failures includes temperature failures, speed failures, vibration failures, and load failures. The cumulative number of failures includes the cumulative number of temperature failures, speed failures, vibration failures, and load failures. 052: Input any one of the statistical failure counts, any one of the cumulative failure counts, and the usage duration into a preset linear regression model for training, so as to obtain multiple pre-trained linear regression models.

[0063] Please combine Figure 3 In the life prediction device 100 of this application embodiment, the processor 130 is also used to execute the life prediction methods in 051 and 052. That is, the processor 130 is also used to: obtain training parameters, which include the statistical number of failures of key components in each historical time window and the cumulative number of failures in the used time before any time window. The statistical number of failures includes the number of temperature failures, the number of speed failures, the number of vibration failures and the number of load failures. The cumulative number of failures includes the cumulative number of temperature failures, the cumulative number of speed failures, the cumulative number of vibration failures and the cumulative number of load failures; and input any one of the statistical number of failures, any one of the cumulative number of failures and the used time into a preset linear regression model for training to obtain multiple pre-trained linear regression models.

[0064] Specifically, in the above embodiment, the processor 130 can use the statistical number of failures of key components of the machine in each historical time window, and the cumulative number of failures within the usage time before any time window, as training parameters for training a linear regression model. The preset model is an initial model with a preset structure for training. The processor 130 can input the acquired training parameters into the preset model to train the preset model and obtain a linear regression model.

[0065] Specifically, the processor 130 can input the number of temperature faults from the statistical fault count, the cumulative number of temperature faults from the cumulative fault count, and the usage time into a preset model for training to obtain a linear regression model related to temperature fault prediction; input the number of vibration faults from the statistical fault count, the cumulative number of vibration faults from the cumulative fault count, and the usage time into a preset model for training to obtain a linear regression model related to vibration fault prediction; input the number of speed faults from the statistical fault count, the cumulative number of speed faults from the cumulative fault count, and the usage time into a preset model for training to obtain a linear regression model related to speed fault prediction; and input the number of load faults from the statistical fault count, the cumulative number of load faults from the cumulative fault count, and the usage time into a preset model for training to obtain a linear regression model related to load fault prediction.

[0066] Thus, the lifespan prediction method of this application can obtain four independent linear regression models through training parameters. In subsequent applications, such as in step 03, the processor 130 can input the total used time and the cumulative number of failures corresponding to the linear regression model into the linear regression model to obtain the first predicted number of failures of the key component in the first prediction time window immediately following the current time window, and update the cumulative number of failures according to the first predicted number of failures to obtain the first predicted cumulative number of failures. The linear regression model can predict the second predicted number of failures in the next prediction time window based on the first predicted cumulative number of failures and the total used time... and repeat the above steps; for example, the processor 130 can input the total used time... The number of long and cumulative temperature failures is input into a linear regression model related to temperature prediction to obtain the first predicted number of temperature failures of the key component within the first prediction time window. The cumulative number of temperature failures is updated based on the first predicted number of temperature failures to obtain the first predicted cumulative number of temperature failures. In step 04, the processor 130 can input the predicted number of failures and the predicted cumulative number of failures corresponding to each prediction time window obtained by all linear regression models into the decision tree model for judgment. When the judgment result output by the decision tree model is a replacement instruction, the number of loop iterations completed at this time is obtained, and the remaining service life of the key component is obtained based on the number of loop iterations and the length of a single prediction time window.

[0067] In some implementations, 052: any one of the statistical failure counts, any one of the cumulative failure counts, and the used duration are input into a preset linear regression model for training, to obtain multiple pre-trained linear regression models, including: 0521: Based on minimizing the error, any one of the statistical failure counts, any one of the cumulative failure counts, and the duration of use are processed to obtain the intercept (β0) and regression coefficients (β1, β2); and 0522: Input the intercept and regression coefficients into the preset linear regression model to obtain the pre-trained linear regression model.

[0068] Please combine Figure 3 The processor 130 in the life prediction device 100 of this application embodiment is also used to execute the life prediction methods in 0521 and 0522. That is, the processor 130 is also used to: process any one of the statistical failure counts, any one of the cumulative failure counts, and the usage time based on minimizing the error to obtain the intercept (β0) and regression coefficients (β1, β2); and input the intercept and regression coefficients into a preset linear regression model to obtain a pre-trained linear regression model.

[0069] It should be noted that, in the above embodiments, the preset model can be: ; in, For the intercept term, , The coefficients of the two independent variables are ε, where ε represents the error term (or disturbance term).

[0070] Specifically, in the life prediction method of this application, the linear regression model can learn the linear mapping relationship between key parameters and the number of failures by minimizing errors based on historical failure data. Specifically, using historical failure data, life curves of key components can be plotted to reveal the degradation process of these components; based on life curve analysis, the duration of use and the cumulative number of failures are selected as key parameters (X1 and X2), and the number of failures on the second day is used as the output (Y), thereby obtaining the linear regression model.

[0071] For example, processor 130 can acquire historical data of key components over the past 100 days and use it as training parameters to train a linear regression model. For instance, processor 130 can input the statistical failure data (Y) of day 81, the cumulative failure data (X2) of the previous 80 days, and the usage time of 80 days (X1) into a preset model to obtain the intercept (β0) and regression coefficients (β1, β2). Based on the obtained intercept (β0) and regression coefficients (β1, β2), a linear regression model is obtained. ε represents the part that the linear regression model fails to explain, including random noise, unexplained variables, measurement errors, etc., which can generally be estimated through residuals (i.e., the difference between the actual value and the model prediction).

[0072] It should be noted that in some embodiments, by injecting the latest fault data daily, the model parameters are triggered to iterate and adapt to parameter drift caused by equipment aging. For example, the changes in thermal deformation parameters after long-term operation of the spindle can be captured in a timely manner, thereby improving the accuracy of subsequent life prediction.

[0073] In some implementations, the lifetime prediction method further includes obtaining a pre-defined decision tree model in the following manner: 061: Obtain multiple training features, including the statistical number of failures of key components in each historical time window, and the cumulative number of failures within the cumulative used time. The statistical number of failures includes the number of temperature failures, speed failures, vibration failures, and load failures. The cumulative number of failures includes the cumulative number of temperature failures, speed failures, vibration failures, and load failures. 062: Calculate the information gain of multiple training features using a preset algorithm; 063: Select the training feature with the largest information gain as the splitting node; and 064: Recursively split the training features of each split node, and return to continue executing the step of calculating the information gain of multiple training features using a preset algorithm, until the recursive splitting stops after the recursion termination condition is met, so as to obtain the preset decision tree model. The recursive termination condition is as follows: when the depth of the recursive split reaches the maximum depth threshold determined by cross-validation or the number of training samples contained in the current split node is lower than the preset minimum number of samples threshold, the splitting stops and the current split node is marked as a leaf node. The output of the leaf node is "replace instruction" or "do not replace instruction", thus obtaining the preset decision tree model.

[0074] Please combine Figure 3 The processor 130 in the life prediction device 100 of this application embodiment is also used to execute the life prediction methods in 061, 062, 063 and 064. That is, the processor 130 is also used to: acquire multiple training features, the training features including the statistical number of failures of key components in each historical time window, and the cumulative number of failures in the cumulative used time. The statistical number of failures includes the number of temperature failures, the number of speed failures, the number of vibration failures and the number of load failures. The cumulative number of failures includes the cumulative number of temperature failures, the cumulative number of speed failures, the cumulative number of vibration failures and the cumulative number of load failures. The processor 130 is also used to: calculate the information gain of the multiple training features using a preset algorithm. The training feature with the largest information gain is selected as the splitting node; and the training features of each splitting node are recursively split, and the process continues to execute the steps of calculating the information gain of multiple training features using a preset algorithm, until the recursive splitting stops when the recursive termination condition is met, so as to obtain the preset decision tree model; wherein, the recursive termination condition is: when the depth of the recursive splitting reaches the maximum depth threshold determined by cross-validation or the number of training samples contained in the current splitting node is lower than the preset minimum number of samples threshold, the splitting stops and the current splitting node is marked as a leaf node, and the output of the leaf node is "replace instruction" or "do not replace instruction", thus obtaining the preset decision tree model.

[0075] Specifically, in the above embodiments, the processor 130 can acquire multiple training features and calculate the information gain of each training feature using a preset algorithm. For example, for each training feature, the information gain is calculated using the information gain formula, and the training feature with the largest information gain is selected as the splitting node. Subsequently, the processor 130 can recursively split the training features of each splitting node and return to step 062 until all samples (training features) are assigned to leaf nodes, and each leaf node corresponds to a type. That is, the processor 130 can recursively split the training features of each splitting node and return to step 062 until the recursion stops after the recursion termination condition is met, so as to obtain the preset decision tree model. It should be noted that in some embodiments, the recursion termination condition includes at least one of the following: the maximum depth is greater than the maximum depth threshold 10, and the number of training samples in the node is less than the preset minimum sample number threshold 500.

[0076] Optionally, information gain = ; Where S is the entire sample set, and in some embodiments of this application, S can be 10000; T is the value of the target variable in S, and in some embodiments of this application, T={0,1}; St is a subset of samples in S where the target variable takes the value t.

[0077] Optionally, the processor 130 can input the predicted number of failures obtained by iteratively traversing multiple linear regression models according to the prediction time window, and the predicted cumulative number of failures obtained by the processor 130 based on the predicted number of failures and the cumulative number of failures, into the trained decision tree model. If the decision tree model outputs a replacement instruction, the recursion is terminated, the number of iterations of the linear regression model is obtained, and the remaining lifespan days of the critical component are obtained based on the number of iterations and the length of the prediction time window. For example, if the length of the prediction time window is 1 day and the number of iterations is 3, the remaining lifespan can be 3 days.

[0078] In some implementations, the lifespan prediction method further includes assessing the overall operating status of the machine tool based on working data and remaining service life. Specifically, by constructing a data integration platform for unified management and processing of remaining service life and other key performance indicator data of critical components, the current data processing mechanism is enabled to perform real-time analysis of collected data such as temperature, rotational speed, vibration, and load, and combine this with the remaining service life to assess the overall operating status of the machine tool.

[0079] In other embodiments, the life prediction method further includes: pushing early warning information when the remaining life of a critical component is lower than a preset life threshold.

[0080] Specifically, when the remaining service life is lower than a preset service life threshold, warning messages can be sent to the user through multiple channels to ensure that the user receives the remaining service life information in a timely manner, reminding them to formulate maintenance plans and replace key components promptly to avoid potential production interruptions and equipment failures. It should be noted that in some embodiments, the push notification methods include, but are not limited to, email, SMS, and mobile application push notifications.

[0081] Please see Figure 3 , Figure 3 This is a schematic diagram of the structure of a life prediction device 100 for a key component of a machine tool, provided in one embodiment of this application. It is understood that... Figure 3 The illustrated structure is merely an example of the life prediction device 100 and does not constitute a specific limitation on the life prediction device 100. The life prediction device 100 can be a device with an application installed, such as a computer, mobile phone, tablet computer, or personal digital assistant (PDA). The life prediction device 100 may include more or fewer components than illustrated, or combine certain components, or different components. For example, the life prediction device 100 may also include input / output devices, network access devices, buses, etc.

[0082] The life prediction device 100 provided in this application includes a memory 110 and a processor 130. It should be noted that the life prediction device 100 in this embodiment is the same as the life prediction device 100 in the above embodiments. Therefore, the explanations and descriptions of the life prediction device 100 in the previous embodiments also apply to the life prediction device 100 in this embodiment, and similarly, the explanations and descriptions of the life prediction device 100 in this embodiment also apply to the life prediction device 100 in the previous embodiments.

[0083] Specifically, memory 110 may include one or more random access memories (RAM) and one or more non-volatile memories (NVM).

[0084] The random access memory can be directly read and written by the processor 130. It can be used to store executable programs of other running programs, as well as user and application data. The random access memory can include, but is not limited to, static random-access memory (SRAM), dynamic random-access memory (DRAM), synchronous dynamic random-access memory (SDRAM), and double data rate synchronous dynamic random-access memory (DDRSDRAM).

[0085] Non-volatile memory can also store executable programs and user and application data, and can be pre-loaded into random access memory for direct reading and writing by the processor 130. Non-volatile memory can include disk storage devices and flash memory. For example, flash memory can be NandFlash.

[0086] The memory 110 stores a computer program configured to be executed by the processor 130 to implement the life prediction method in any of the above embodiments. The computer program may include at least one instruction, which, when executed by the processor 130, can implement the life prediction method for key machine components in any of the above embodiments.

[0087] In other embodiments, the lifetime prediction device 100 may also include an external memory interface for connecting to an external memory to expand the storage capacity of the lifetime prediction device 100.

[0088] The processor 130 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or a processor, or any conventional processor.

[0089] The processor 130 provides computing and control capabilities, for example, the processor 130 is used to execute instructions stored in the memory 110 to implement the life prediction method for key components of the machine tool in any of the above embodiments.

[0090] It should be noted that the explanation of the life prediction method for key components of the machine tool in the above embodiments also applies to the life prediction device 100 in this embodiment, and will not be explained further here.

[0091] In the life prediction device 100 of this application embodiment, statistical results are obtained based on the current and historical data of key components of each machine. The statistical results include the types of failures that occur in the key components, the total used time from the first use of the key component to the end of the current time window, and the cumulative number of failures within the total used time. A pre-trained linear regression model is used to obtain the first predicted number of failures of the key component within the first prediction time window immediately following the current time window. The first predicted number of failures is added to the cumulative number of failures within the total used time to obtain the first predicted cumulative number of failures. Starting from the first prediction time window, multiple consecutive predictions are performed for each subsequent prediction time window. The prediction time window is iterated repeatedly to obtain the predicted number of failures and the predicted cumulative number of failures for each prediction time window. The predicted number of failures and the predicted cumulative number of failures for each prediction time window are used as a set of input features and are sequentially input into a preset decision tree model for judgment according to the order of the prediction time windows. When the decision tree model outputs a replacement instruction, the number of iterations completed at this time is obtained. Finally, based on the number of iterations and the length of a single prediction time window, the remaining service life of the key components is obtained. This can provide data support for failure prevention, realize predictive maintenance, avoid over-maintenance and sudden failure downtime, reduce maintenance costs, and improve the stability and reliability of machine operation.

[0092] Please combine Figure 3 The computer-readable storage medium of the embodiments of this application stores a computer program, which, when executed by the processor 130 in the life prediction device 100, implements the life prediction method for key machine components as described in any of the above embodiments.

[0093] The computer-readable storage medium can be the internal memory of the life prediction device 100 in the above embodiments, such as the hard disk or memory of the life prediction device 100. The computer-readable storage medium can also be an external storage device of the life prediction device 100, such as a plug-in hard disk, smart media card (SMC), flash card, etc. equipped on the life prediction device 100.

[0094] In some embodiments, a computer-readable storage medium may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program required for at least one function, etc.; and the data storage area may store data created based on the use of the lifetime prediction device 100, etc.

[0095] It should be noted that the explanations of the life prediction method and life prediction device 100 for the key components of the machine tool in the foregoing embodiments also apply to the computer-readable storage medium of this embodiment, and will not be elaborated here.

[0096] In the computer-readable storage medium of this application embodiment, statistical results are obtained based on current and historical data of key components of each machine. The statistical results include the types of failures occurring in the key components, the total used time from the first use of the key component to the end of the current time window, and the cumulative number of failures within the total used time. A pre-trained linear regression model is used to obtain the first predicted number of failures for the key component within the first predicted time window immediately following the current time window. The first predicted number of failures is added to the cumulative number of failures within the total used time to obtain the first predicted cumulative number of failures. Starting from the first predicted time window, multiple consecutive predicted failures are then processed. The prediction time window is iterated repeatedly to obtain the predicted number of failures and the predicted cumulative number of failures for each prediction time window. The predicted number of failures and the predicted cumulative number of failures for each prediction time window are used as a set of input features and are sequentially input into a preset decision tree model for judgment according to the order of the prediction time windows. When the decision tree model outputs a replacement instruction, the number of iterations completed at this time is obtained. Finally, based on the number of iterations and the length of a single prediction time window, the remaining service life of the key components is obtained. This can provide data support for failure prevention, realize predictive maintenance, avoid over-maintenance and sudden failure downtime, reduce maintenance costs, and improve the stability and reliability of machine operation.

[0097] The computer program product of this application includes a computer program that, when executed by a processor, implements the life prediction method for key machine components of any of the above embodiments.

[0098] In the computer program product of this application, statistical results are obtained based on the current and historical data of the key components of each machine. The statistical results include the types of failures that occurred in the key components, the total usage time from the first time the key components were used to the end of the current time window, and the cumulative number of failures within the total usage time. A pre-trained linear regression model is then used to obtain the first predicted number of failures for the key components within the first predicted time window immediately following the current time window. The first predicted number of failures is added to the cumulative number of failures within the total usage time to obtain the first predicted cumulative number of failures. Starting from the first predicted time window, multiple consecutive predictions are made for the failures after the first predicted time window. The time window is iterated repeatedly to obtain the predicted number of failures and the predicted cumulative number of failures for each prediction time window. The predicted number of failures and the predicted cumulative number of failures for each prediction time window are used as a set of input features and are sequentially input into a preset decision tree model for judgment according to the order of the prediction time windows. When the decision tree model outputs a replacement instruction, the number of iterations completed at this time is obtained. Finally, based on the number of iterations and the length of a single prediction time window, the remaining service life of the key components is obtained. This can provide data support for failure prevention, realize predictive maintenance, avoid over-maintenance and sudden failure downtime, reduce maintenance costs, and improve the stability and reliability of machine operation.

[0099] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0100] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the function involved, as will be understood by those skilled in the art to which embodiments of this application pertain.

[0101] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer storage medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, a computer storage medium can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer storage media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer storage medium could be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in the computer memory.

[0102] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0103] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer storage medium, and when executed, it includes one or a combination of the steps of the method embodiments. Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc.

[0104] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for predicting the lifetime of a critical component of a machine, characterized in that, The method comprises the following steps: obtaining working data of a key component of the machine, the working data comprising current data and historical data; obtaining a statistical result according to the current data and the historical data, the statistical result comprising a total used time length of a failure type of the key component from a first time when the key component is first used to a current time window end time, and a cumulative failure number in the total used time length, wherein the current time window is a fixed length and continuous time period; inputting the total used time length and the cumulative failure number into a pre-trained linear regression model to output a first predicted failure number of the key component in a first prediction time window next to the current time window, adding the first predicted failure number to the cumulative failure number to obtain a first predicted cumulative failure number, wherein the output of the pre-trained linear regression model is determined by at least an intercept, a first regression coefficient related to the total used time length, and a second regression coefficient related to the cumulative failure number, and when the failure type is multiple, the pre-trained linear regression model can be set for different failure types respectively; and starting from the first prediction time window, performing a loop iteration on a plurality of continuous prediction time windows after the first prediction time window to obtain a predicted failure number and a predicted cumulative failure number corresponding to each prediction time window; inputting the predicted failure number and the predicted cumulative failure number corresponding to each prediction time window as a group of input features into a preset decision tree model in turn according to the order of the prediction time windows for judgment; wherein when the failure type is multiple, the corresponding predicted failure number and predicted cumulative failure number are also multiple and correspond one by one; when the decision tree model outputs a replacement instruction, obtaining the number of times the linear regression model has been called in the loop iteration process on the plurality of continuous prediction time windows, taking the number of times as a loop iteration number, and obtaining a remaining service life of the key component according to the loop iteration number and the length of a single prediction time window.

2. The life prediction method according to claim 1, characterized by, The working data of the key component comprises: collecting current data and historical data of the key component, the current data comprising a current use time length of a failure type of the key component from a current time window start time to a current time window end time, and a current failure number occurring in the current use time length; the historical data comprising a historical used time length of the failure type of the key component from a first time when the key component is first used to a current time window start time, and a historical failure number occurring in the historical used time length; performing compression processing on the current data to obtain current compressed data; performing buffered transmission on the current compressed data using Apache Kafka, and removing abnormal data in the current compressed data using a clustering analysis algorithm to obtain target data; and Merge the target data and the historical data together as working data of the key component.

3. The life prediction method according to claim 2, characterized by, The statistical result is obtained according to the current data and the historical data, including: The failure type is multiple, and according to the current data and the historical data, all failure types of the key component occurring in the current time window and the failure mode corresponding to each failure type are obtained; Using the FMEA method, the failure score of each failure mode corresponding to the failure type is calculated; The failure mode with a failure score not less than a preset score threshold is screened from all the failure modes, and the failure mode with a failure score not less than a preset score threshold is taken as a key failure mode; According to the working data, the failure number of the failure type corresponding to the key failure mode in the current time window of the key component is obtained as a statistical failure number; and According to the statistical failure number of the failure type corresponding to the key failure mode and the historical failure number of the same failure type, the statistical result is obtained, including the cumulative failure number of each failure type corresponding to the key failure mode in the current service time.

4. The life prediction method according to claim 3, characterized by, The data type of the working data includes temperature, speed, vibration and load; Correspondingly, the failure type includes temperature failure, speed failure, vibration failure and load failure; The historical failure number includes historical temperature failure number, historical speed failure number, historical vibration failure number and historical load failure number; The statistical failure number includes temperature failure number, speed failure number, vibration failure number and load failure number; The cumulative failure number includes temperature cumulative failure number, speed cumulative failure number, vibration cumulative failure number and load cumulative failure number.

5. The life prediction method of claim 1, wherein: The failure type is multiple, and the pre-trained linear regression model is multiple, each of which corresponds to a failure type; In each iteration of the cyclic iteration, for each failure type, the cumulative failure number and the current service time of the failure type are input into the corresponding linear regression model to obtain the predicted failure number of the failure type in the prediction time window, and the predicted cumulative failure number of the failure type is updated; The predicted failure number and the predicted cumulative failure number corresponding to each prediction time window are taken as a group of input features, which are input into a preset decision tree model in the order of prediction time window; when the decision result output by the decision tree model is a replacement instruction, the number of times the linear regression model has been called in the process of cyclic iteration of the multiple continuous prediction time windows is obtained, the number of times is taken as the cyclic iteration number, and the remaining service life of the key component is obtained according to the cyclic iteration number and the length of a single prediction time window; including: The predicted failure times of each of the failure types and the predicted cumulative failure times of each of the failure types are respectively taken as a group of input features, and are sequentially input into a preset decision tree model in the order of prediction time windows for judgment; When the decision result output by the decision tree model is a replacement instruction, the number of times the linear regression model has been called for the failure type in the process of cyclic iteration on the plurality of continuous prediction time windows is obtained, the number of times is taken as a cyclic iteration number, and the remaining service life of the key component corresponding to the failure type is obtained according to the cyclic iteration number and the length of a single prediction time window.

6. The life prediction method according to claim 1, characterized by, The life prediction method further comprises: obtaining training parameters, the training parameters including statistical failure times of the key component in each historical time window, and cumulative failure times in the used time length before any of the historical time windows, the statistical failure times including temperature failure times, rotation speed failure times, vibration failure times and load failure times, and the cumulative failure times including temperature cumulative failure times, rotation speed cumulative failure times, vibration cumulative failure times and load cumulative failure times; and inputting any of the statistical failure times, any of the cumulative failure times and the used time length into a preset linear regression model for training, to obtain a plurality of pre-trained linear regression models.

7. The life prediction method according to claim 6, characterized by, The inputting any of the statistical failure times, any of the cumulative failure times and the used time length into a preset linear regression model for training, to obtain a plurality of pre-trained linear regression models, comprises: processing any of the statistical failure times, any of the cumulative failure times and the used time length based on minimizing error, to obtain an intercept and a regression coefficient; and inputting the intercept and the regression coefficient into a preset linear regression model, to obtain a pre-trained linear regression model.

8. The life prediction method according to claim 1, characterized by, The life prediction method further comprises that the preset decision tree model is obtained by: obtaining a plurality of training features, the training features including statistical failure times of the key component in each historical time window, and cumulative failure times in the cumulative used time length, the statistical failure times including temperature failure times, rotation speed failure times, vibration failure times and load failure times, and the cumulative failure times including temperature cumulative failure times, rotation speed cumulative failure times, vibration cumulative failure times and load cumulative failure times; calculating information gains of a plurality of the training features respectively by using a preset algorithm; selecting the training feature with the largest information gain as a split node; and recursively splitting the training features of each of the split nodes, and returning to continue executing the step of calculating information gains of a plurality of the training features by using a preset algorithm, until recursive splitting is stopped after a recursive termination condition is met, to obtain a preset decision tree model; The recursive termination condition is that when the depth of recursive splitting reaches a maximum depth threshold determined by cross-validation or the number of training samples contained in the current split node is lower than a preset minimum sample number threshold, the splitting is stopped and the current split node is marked as a leaf node, and the output of the leaf node is "replacement instruction" or "no replacement instruction", thereby obtaining the preset decision tree model.

9. A device for predicting the lifetime of a key component of a machine, characterized in that, The method comprises the following steps: a collection layer configured to obtain working data of a key component of the machine, the working data including current data and historical data; a processing layer configured to obtain statistical results from the current data and the historical data, the statistical results including a total used duration of a failure type of the key component from a first use time of the key component to an end time of a current time window and an accumulated failure number within the total used duration, wherein the current time window is a fixed-length and continuous time period; a first decision layer configured to input the total used duration and the accumulated failure number into a pre-trained linear regression model to output a first predicted failure number of the key component in a first prediction time window immediately after the current time window, and add the first predicted failure number to the accumulated failure number within the total used duration to obtain a first predicted accumulated failure number, wherein the output of the pre-trained linear regression model is determined by at least an intercept, a first regression coefficient related to the total used duration, and a second regression coefficient related to the accumulated failure number within the total used duration, and the pre-trained linear regression model can be set for different failure types when there are multiple failure types; and a second decision layer configured to perform cyclic iteration on multiple continuous prediction time windows after the first prediction time window from the first prediction time window to obtain a predicted failure number and a predicted accumulated failure number corresponding to each prediction time window, input the predicted failure number and the predicted accumulated failure number corresponding to each prediction time window as a group of input features into a preset decision tree model in a sequential order of the prediction time windows for judgment, and when the judgment result output by the decision tree model is a replacement instruction, obtain a number of times the linear regression model has been called in the process of cyclic iteration on the multiple continuous prediction time windows, take the number of times as a cyclic iteration number, and obtain a remaining service life of the key component according to the cyclic iteration number and a length of a single prediction time window.

10. A device for predicting the lifetime of a critical component of a machine, comprising a memory and a processor, the memory being for storing instructions, characterized in that, The instructions stored in the memory are executed by the processor to implement the service life prediction method of any one of claims 1-8.

Images