Industrial edge machine room hard disk fault prediction method based on IIOT
By using IIoT to collect data in hard disk failure prediction and using SVR model, combined with FARM indicators and feature selection, the problems of low prediction accuracy and poor adaptability in the existing technology are solved, and higher prediction accuracy and reliability are achieved reduces false alarms and missed rates.
Patent Information
- Application Number
- CN202410920150.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-10
- Publication Date
- 2024-11-12
AI Technical Summary
The existing hard disk failure prediction technology has high computational complexity when processing large-scale data, and it is difficult to capture the long-term dependencies of timing data. It also has poor prediction accuracy for hard disks of different brands and models, which has problems such as low prediction accuracy and difficulty in selecting features.
Using IIoT-based method, the hard disk operating environment parameters and FARM index data are collected through the Internet of Things sensor, the hard disk failure prediction is used using the SVR model, the relevant characteristics are selected, the kernel functions and parameters are optimized to improve the prediction accuracy, which is divided into faults, data exceptions and In normal three types of states, adjust model parameters to reduce false alarm rate and false alarm rate.
Improve the accuracy and flexibility of hard disk failure prediction, enhance the generalization ability of the model, reduce the false alarm rate and missed alarm rate, improve the reliability and adaptability of the prediction, can better deal with different hard disk failures, and reduce Downtime and data loss.
Smart Images

Figure CN118939487A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of hard disk failure prediction, and in particular to an IIoT-based industrial edge computer room hard disk failure prediction method. Background Art
[0002] With the rapid development of the industrial Internet, the speed at which industrial enterprises move to the cloud and platform is gradually accelerating. More and more industrial production data are being collected and stored. Industrial data plays a very important role in guiding industrial production processes and recording industrial production status. Once data is damaged or lost, it may affect the entire production and manufacturing process. Hard disks are important equipment for industrial data storage. The stable operation of hard disks helps to improve the security and stability of industrial data storage.
[0003] With the continuous growth of data storage capacity, hard disk failure has become the norm rather than the exception. Hard disk failure prediction is mainly based on the learning and training of historical failure data of hard disk devices to obtain a hard disk failure prediction model, predict hard disk failures, and conduct early warning and prevention, so that data can be stored more reliably. At present, many researchers have carried out research on hard disk failure prediction technology. For example, patent CN109240867A discloses a hard disk failure prediction method, which uses SMART values as collected data, inputs SMART values into the trained deep belief network DBN model, and obtains the hard disk failure warning time; patent CN107392320A discloses a method for predicting hard disk failures using machine learning, which uses a random forest algorithm to train and model historical data, generate a fault prediction model, and improve the fault prediction rate.
[0004] However, the existing hard disk failure prediction technology similar to the above application still has the following shortcomings:
[0005] Deep belief networks, or DBNs, can extract high-level feature representations from data, but they may encounter problems with high computational complexity and long training time when processing large-scale data. In addition, although DBNs have powerful feature extraction capabilities, they may not be flexible enough in processing time series data and cannot fully capture long-term dependencies in time series data. Although the random forest algorithm has high accuracy and stability, it may encounter problems with feature selection when processing high-dimensional data, and it is not its strong point in processing time series data. In addition, the random forest algorithm also needs to be cautious in handling missing values and outliers, otherwise it may affect the performance of the model. In summary, existing prediction technologies generally have problems such as low prediction accuracy and differences in prediction accuracy for hard drives of different brands and models, which is not conducive to the application and promotion of hard drive failure prediction technology.
[0006] Therefore, there is an urgent need to improve this shortcoming. The present invention studies and improves the existing technology and its shortcomings, and provides an industrial edge computer room hard disk failure prediction method based on IIoT. Summary of the invention
[0007] The purpose of the present invention is to provide an IIoT-based industrial edge computer room hard disk failure prediction method to solve the problems raised in the above background technology.
[0008] To achieve the above object, the present invention provides the following technical solution: a method for predicting hard disk failure in an industrial edge computer room based on IIoT, comprising the following steps:
[0009] S1. Data collection and preprocessing: IoT sensors are deployed to collect various environmental parameters of the hard disk during operation, and FARM indicators are used to obtain key reliability indicators from the hard disk itself or its operating environment through specific data collection software or equipment. The environmental parameter data collected by the IoT sensors and the FARM indicator data are then integrated.
[0010] S2, Feature selection: Select the feature subset that is most relevant to hard disk failure prediction from the features, and select the most important features as the input of the model;
[0011] S3, model construction:
[0012] The SVR model is used to fit the hard disk operation status data. The performance parameters x1, x2, x3, ..., xn that are closely related to the hard disk failure are selected through step S2. The hard disk status is recorded as x = (x1, x2, x3, ..., xn), and the hard disk status is marked as three types, namely, the failure state y = 0, the slight data abnormality but still able to operate normally y = 1, and the normal state y = 2;
[0013] The SVR model is used to train the hard disk operation status data level (x, y) to train the fitting function: y = ω T +b, and use the insensitive loss function threshold ε to control the function boundary, stipulating that the predicted hard disk status between f(x)+ε and f(x)-ε is a normal hard disk, the hard disk status between f(x)+ε1 to f(x)+ε2, and f(x)-ε1 to f(x)–ε2 is a hard disk with abnormal data but can still operate normally, and the hard disk status exceeds f(x)-ε2 or is lower than f(x)-ε2 is a faulty hard disk. As long as the predicted value is between the corresponding dotted lines, it means the prediction is correct;
[0014] Maximize the interval band and minimize the loss to determine the values of ω and b and improve the accuracy of prediction. After introducing the slack variable ξi≥0, the optimization problem of support vector machine regression is expressed as follows:
[0015]
[0016] st y i -ω T x i -b≤ε+ξ i
[0017] ω T x i +by i ≤ε+ξ i *
[0018] ξ i ≥0,
[0019] S4, parameter tuning: adjust the parameters within the SVR model, which control the complexity and generalization ability of the model, including:
[0020] C, the penalty coefficient: used to control the model's tolerance to errors;
[0021] gamma, the coefficient of the kernel function: for SVR using the radial basis function kernel, it is used to determine the range of influence between data points;
[0022] ε, the threshold in the insensitive loss function: used to define the width of the region where the regression function is insensitive to sample data;
[0023] S5. Model evaluation: Use the test set to evaluate the model, apply the SVR model with the optimal parameters to the test set data, and calculate the detection rate, false alarm rate, and missed alarm rate predicted by the SVR model.
[0024] Furthermore, in step S1, the specific operation of collecting data by the IoT sensor is: in the industrial environment where the hard disk is running, various IoT sensors including temperature sensors, humidity sensors, and vibration sensors are deployed to collect various environmental parameters of the hard disk in real time during operation, such as temperature, humidity, vibration, etc., through the sensors, and the collected data is transmitted to the data collection system via wired or wireless means.
[0025] Furthermore, in step S1, FARM indicator data includes but is not limited to the read and write speed, response time, and error rate of the hard disk, which are used to reflect the health status and potential failure risks of the hard disk. In the current hard disk operation and maintenance practice, the data collection of the hard disk is based on the SMART system. FARM is an improvement on the SMART system. FARM itself is easy to obtain and has a wide range of parameters. More than 120 parameters of FARM include relevant parameters at the lower-level sensor and head level of the hard disk, as well as parameters of usage conditions such as flight altitude, workload, and environment, which can perform comprehensive detection of the hard disk.
[0026] Furthermore, in step S1, data integration realizes unified management and analysis of data from different sources by setting a data interface in the data collection system, and the integrated data will form a comprehensive hard disk operation data set to provide data support for subsequent hard disk failure prediction.
[0027] Furthermore, in step S2, the specific operation of feature selection is as follows:
[0028] S21. Use the correlation analysis method in statistics to measure the linear or nonlinear relationship between features and between features and hard disk failures;
[0029] S22. Draw a correlation heat map or matrix to intuitively show the strength of the correlation between features;
[0030] S23. According to the size of the correlation coefficient, select features that are highly correlated with hard disk failure.
[0031] Furthermore, in the SVR model constructed in step S3, the kernel function determines the mapping method of data in high-dimensional space. The selection range of the kernel function includes linear kernel, polynomial kernel, radial basis function kernel and Sigmoid kernel. For the hard disk failure prediction problem, the data has nonlinear characteristics, and the radial basis function kernel is preferentially selected. The radial basis function kernel is suitable for processing nonlinear relationships and is relatively insensitive to the selection of parameters.
[0032] Furthermore, in step S3, the goal of hard disk failure prediction is to increase the detection rate and reduce the false alarm rate and missed alarm rate.
[0033] FDR: Fault detection rate or recall rate, which indicates the proportion of all faulty disks that are detected.
[0034] False alarm rate (FAR): also known as false alarm rate or false detection rate, which indicates the proportion of all normal disks that are predicted to be faulty disks;
[0035] Missing alarm rate (MAR): also known as missed detection rate or missed warning rate, which indicates the proportion of faulty disks predicted to be normal disks.
[0036] Furthermore, in step S4, the tuning logic of the C parameter is: the larger the C value, the better the model fits the training data, but too large a C value will lead to overfitting. Conversely, the smaller the C value, the smaller the penalty for error imposed by the model, and too small a C value will lead to underfitting. Adjusting the value of C helps balance the complexity and generalization ability of the model.
[0037] Furthermore, in step S4, the tuning logic of the gamma parameter is: the larger the gamma value, the smaller the influence range of each data point, and the model becomes more complex and prone to overfitting; the smaller the gamma value, the simpler the model.
[0038] Furthermore, in step S4, the tuning logic of the ε parameter is: when the difference between the predicted value and the actual value is less than ε, the model will not penalize these differences. Conversely, when the difference between the predicted value and the actual value is greater than or equal to ε, the model will penalize these differences. Adjusting the value of ε helps control the sensitivity of the model to errors.
[0039] The present invention provides an IIoT-based industrial edge computer room hard disk failure prediction method, which has the following beneficial effects:
[0040] The present invention proposes a method for predicting industrial data storage hard disk failures based on SVR, which uses FARM to collect hard disk data, thereby improving the convenience and reliability of data collection. On this basis, data with a high correlation with hard disk failure prediction is selected as model data, and the actual data relationship is taken into account during the model establishment process. The RBF kernel is used to improve the accuracy of model training; in addition, by dividing the hard disk status into three categories: failure, data abnormality but still normal operation, and normal, the flexibility and accuracy of hard disk failure prediction are further improved, and users can divide the hard disk status into multiple levels according to the needs of failure prediction; at the same time, it is clarified that the goal of model training is to improve the detection rate and reduce the missed alarm rate and false alarm rate, so that the model has stronger adaptability and can better cope with various hard disk failure prediction activities. The advantages of the present invention are mainly reflected in four aspects:
[0041] First, it improves prediction accuracy: Since hard disk failures often involve multiple complex factors, traditional methods may find it difficult to accurately capture these relationships. The SVR model can more comprehensively consider various factors by finding the optimal hyperplane in high-dimensional space, thereby more accurately predicting hard disk failures. By selecting appropriate kernel functions and parameter tuning, the SVR model can fit the data more accurately, thereby improving prediction accuracy.
[0042] The second is to enhance the generalization ability: In practical applications, hard disk failure data often contains noise and outliers, and the distribution of new data may also differ from the training data. The generalization ability of the SVR model enables it to maintain stable prediction performance under these conditions. The SVR model prevents overfitting by maximizing the width of the boundary band during training, which enables the model to maintain good prediction performance on new data. In addition, the SVR model is highly robust to outliers and noise, and can reduce the impact of these factors on the prediction results.
[0043] The third is to reduce the false alarm rate and the missed alarm rate: False alarm and missed alarm are the issues that need to be focused on in hard disk failure prediction. Too high a false alarm rate may lead to unnecessary maintenance and replacement costs, while a missed alarm may lead to data loss and system crash. The SVR model can effectively reduce the false alarm rate and missed alarm rate and improve the reliability of the prediction by improving the accuracy of the prediction.
[0044] Fourth, improve decision-making efficiency: In industrial applications, rapid response to hard disk failure warnings is crucial to ensuring production efficiency and data security. The efficient prediction capability of the SVR model enables managers to take quick measures to avoid potential risks and losses. Hard disk failure prediction based on the SVR model can provide timely and accurate warning information to help managers make quick and wise decisions, which helps reduce downtime and data loss caused by hard disk failures and improve system availability and stability. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] Figure 1 This is a schematic diagram of the operation flow of an IIoT-based industrial edge computer room hard disk failure prediction method of the present invention;
[0046] Figure 2 This is a schematic diagram of a fitting function of an IIoT-based industrial edge computer room hard disk failure prediction method of the present invention. DETAILED DESCRIPTION
[0047] The following embodiments of the present invention are described in further detail in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.
[0048] like Figure 1-Figure 2 As shown, a method for predicting hard disk failure in an industrial edge computer room based on IIoT includes the following steps:
[0049] S1. Data collection and preprocessing: IoT sensors are deployed to collect various environmental parameters of the hard disk during operation, and FARM indicators are used to obtain key reliability indicators from the hard disk itself or its operating environment through specific data collection software or equipment. The environmental parameter data collected by the IoT sensors and the FARM indicator data are then integrated.
[0050] The specific operation of IoT sensors to collect data is as follows: deploy various IoT sensors including temperature sensors, humidity sensors, and vibration sensors in the industrial environment where the hard disk is running, collect various environmental parameters of the hard disk in real time through the sensors, such as temperature, humidity, vibration, etc., and transmit the collected data to the data collection system via wired or wireless methods;
[0051] FARM indicator data includes but is not limited to the hard disk read and write speed, response time, and error rate, which are used to reflect the health status and potential failure risks of the hard disk. In the current hard disk operation and maintenance practice, the hard disk data collection is based on the SMART system. FARM is an improvement on the SMART system. FARM itself is easy to obtain and has a wide range of parameters. More than 120 parameters of FARM include the relevant parameters of the hard disk's more basic sensor and head level, as well as the parameters of the use conditions such as flight altitude, workload, and environment, which can conduct a comprehensive test of the hard disk.
[0052] Data integration achieves unified management and analysis of data from different sources by setting up data interfaces in the data collection system. The integrated data will form a comprehensive hard disk operation data set, providing data support for subsequent hard disk failure prediction.
[0053] S2. Feature selection: Select the feature subset that is most relevant to hard disk failure prediction from the features, and select the most important features as the input of the model; the specific operations are as follows:
[0054] S21. Use the correlation analysis method in statistics to measure the linear or nonlinear relationship between features and between features and hard disk failures;
[0055] S22. Draw a correlation heat map or matrix to intuitively show the strength of the correlation between features;
[0056] S23, according to the size of the correlation coefficient, filter out features that are highly correlated with hard disk failures;
[0057] S3, model construction:
[0058] The SVR model is used to fit the hard disk operation status data. The performance parameters x1, x2, x3, ..., xn that are closely related to the hard disk failure are selected through step S2. The hard disk status is recorded as x = (x1, x2, x3, ..., xn), and the hard disk status is marked as three types, namely, the failure state y = 0, the slight data abnormality but still able to operate normally y = 1, and the normal state y = 2;
[0059] The SVR model is used to train the hard disk operation status data level (x, y) to train the fitting function: y = ω T +b, and use the insensitive loss function threshold ε to control the function boundary, stipulating that the predicted hard disk status between f(x)+ε and f(x)-ε is a normal hard disk, the hard disk status between f(x)+ε1 to f(x)+ε2, and f(x)-ε1 to f(x)–ε2 is a hard disk with abnormal data but can still operate normally, and the hard disk status exceeds f(x)-ε2 or is lower than f(x)-ε2 is a faulty hard disk. As long as the predicted value is between the corresponding dotted lines, it means the prediction is correct;
[0060] Maximize the interval band and minimize the loss to determine the values of ω and b and improve the accuracy of prediction. After introducing the slack variable ξi≥0, the optimization problem of support vector machine regression is expressed as follows:
[0061]
[0062] st y i -ω T x i -b≤ε+ξ i
[0063]
[0064]
[0065] In the constructed SVR model, the kernel function determines the mapping method of data in high-dimensional space. The kernel function selection range includes linear kernel, polynomial kernel, radial basis function kernel and sigmoid kernel. For the hard disk failure prediction problem, the data has nonlinear characteristics, so the radial basis function kernel is preferred. The radial basis function kernel is suitable for processing nonlinear relationships and is relatively insensitive to the selection of parameters.
[0066] The goal of hard disk failure prediction is to increase the detection rate and reduce the false alarm rate and missed alarm rate.
[0067] FDR: Fault detection rate or recall rate, which indicates the proportion of all faulty disks that are detected.
[0068] False alarm rate (FAR): also known as false alarm rate or false detection rate, which indicates the proportion of all normal disks that are predicted to be faulty disks;
[0069] Missed alarm rate (MAR): also known as missed detection rate or missed warning rate, which indicates the proportion of faulty disks predicted to be normal disks;
[0070] S4. Parameter tuning: Adjust the parameters within the SVR model. The parameters control the complexity and generalization ability of the model, including:
[0071] C, or penalty coefficient, is used to control the tolerance of the model to errors. The tuning logic of the C parameter is: the larger the C value, the better the model fits the training data. However, a too large C value will lead to overfitting. On the contrary, the smaller the C value, the smaller the penalty for errors. A too small C value will lead to underfitting. Adjusting the C value helps balance the complexity and generalization ability of the model.
[0072] Gamma, the coefficient of the kernel function: for SVR using the radial basis function kernel, it is used to determine the influence range between data points; the tuning logic of the gamma parameter is: the larger the gamma value, the smaller the influence range of each data point, and the model becomes more complex and prone to overfitting; the smaller the gamma value, the simpler the model;
[0073] ε, the threshold in the insensitive loss function, is used to define the width of the region where the regression function is insensitive to sample data. The tuning logic of the ε parameter is: when the difference between the predicted value and the actual value is less than ε, the model will not penalize these differences. On the contrary, when the difference between the predicted value and the actual value is greater than or equal to ε, the model will penalize these differences. By adjusting the value of ε, the model helps control the sensitivity of the model to errors.
[0074] S5. Model evaluation: Use the test set to evaluate the model, apply the SVR model with the optimal parameters to the test set data, and calculate the detection rate, false alarm rate, and missed alarm rate predicted by the SVR model.
[0075] The embodiments of the present invention are given for the purpose of illustration and description, and are not intended to be exhaustive or to limit the invention to the disclosed forms. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present invention and to enable those of ordinary skill in the art to understand the present invention and thereby design various embodiments with various modifications suitable for specific uses.
Claims
1. A method for predicting hard disk failure in an industrial edge computer room based on IIoT, characterized in that: The following steps are involved: S1. Data collection and preprocessing: IoT sensors are deployed to collect various environmental parameters of the hard disk during operation, and FARM indicators are used to obtain key reliability indicators from the hard disk itself or its operating environment through specific data collection software or equipment. The environmental parameter data collected by the IoT sensors and the FARM indicator data are then integrated. S2, Feature selection: Select the feature subset that is most relevant to hard disk failure prediction from the features, and select the most important features as the input of the model; S3, model construction: The SVR model is used to fit the hard disk operation status data. The performance parameters x1, x2, x3, ..., xn that are closely related to the hard disk failure are selected through step S2. The hard disk status is recorded as x = (x1, x2, x3, ..., xn), and the hard disk status is marked as three types, namely, the failure state y = 0, the slight data abnormality but still able to operate normally y = 1, and the normal state y = 2; The SVR model is used to train the hard disk operation status data level (x, y) to train the fitting function: y = ω T +b, and use the insensitive loss function threshold ε to control the function boundary, stipulating that the predicted hard disk status between f(x)+ε and f(x)-ε is a normal hard disk, the hard disk status between f(x)+ε1 to f(x)+ε2, and f(x)-ε1 to f(x)–ε2 is a hard disk with abnormal data but can still operate normally, and the hard disk status exceeds f(x)-ε2 or is lower than f(x)-ε2 is a faulty hard disk; Maximize the interval band and minimize the loss to determine the values of ω and b. After introducing the slack variable ξi≥0, the optimization problem of support vector machine regression is expressed as follows: st y i -oh T x i -b≤e+ξ i S4, parameter tuning: adjust the parameters within the SVR model, which control the complexity and generalization ability of the model, including: C: used to control the model's tolerance to errors; gamma: For SVR using radial basis function kernel, it is used to determine the influence range between data points; ε: used to define the width of the region where the regression function is insensitive to sample data; S5. Model evaluation: Use the test set to evaluate the model, apply the SVR model with the optimal parameters to the test set data, and calculate the detection rate, false alarm rate, and missed alarm rate predicted by the SVR model.
2. According to the IIoT-based industrial edge computer room hard disk failure prediction method of claim 1, it is characterized in that: In step S1, the specific operation of collecting data by the IoT sensor is: in the industrial environment where the hard disk is running, various IoT sensors including temperature sensors, humidity sensors, and vibration sensors are deployed, and various environmental parameters of the hard disk are collected in real time through the sensors when the hard disk is running, and the collected data is transmitted to the data collection system via wired or wireless means.
3. The method for predicting hard disk failure in an industrial edge computer room based on IIoT according to claim 1, characterized in that: In step S1, the FARM indicator data includes but is not limited to the hard disk's read and write speed, response time, and error rate, which are used to reflect the hard disk's health status and potential failure risks. In the current hard disk operation and maintenance practice, hard disk data collection is based on the SMART system.
4. The method for predicting hard disk failure in an industrial edge computer room based on IIoT according to claim 1, characterized in that: In step S1, data integration realizes unified management and analysis of data from different sources by setting a data interface in the data collection system, and the integrated data will form a comprehensive hard disk operation data set to provide data support for subsequent hard disk failure prediction.
5. The method for predicting hard disk failure in an industrial edge computer room based on IIoT according to claim 1, characterized in that: In step S2, the specific operation of feature selection is as follows: S21. Use the correlation analysis method in statistics to measure the linear or nonlinear relationship between features and between features and hard disk failures; S22. Draw a correlation heat map or matrix to intuitively show the strength of the correlation between features; S23. According to the size of the correlation coefficient, select features that are highly correlated with hard disk failure.
6. The method for predicting hard disk failure in an industrial edge computer room based on IIoT according to claim 1, characterized in that: In the step S3, in the constructed SVR model, the kernel function determines the mapping method of data in the high-dimensional space. The selection range of the kernel function includes linear kernel, polynomial kernel, radial basis function kernel and sigmoid kernel. For the hard disk failure prediction problem, the radial basis function kernel is preferably selected, and the radial basis function kernel is suitable for processing nonlinear relationships and is relatively insensitive to the selection of parameters.
7. The method for predicting hard disk failure in an industrial edge computer room based on IIoT according to claim 1, characterized in that: In step S3, the goal of hard disk failure prediction is to increase the detection rate and reduce the false alarm rate and missed alarm rate. Detection rate: also known as fault detection rate or recall rate, indicating the proportion of all faulty disks that are detected; False alarm rate: also known as false alarm rate or false detection rate, which indicates the proportion of all normal disks that are predicted to be faulty disks; Missed alarm rate: also known as missed detection rate or missed warning rate, which indicates the proportion of faulty disks that are predicted to be normal disks.
8. The method for predicting hard disk failure in an industrial edge computer room based on IIoT according to claim 1, characterized in that: In step S4, the tuning logic of the C parameter is: the larger the C value, the better the model fits the training data, but too large a C value will lead to overfitting. Conversely, the smaller the C value, the smaller the penalty for error, and too small a C value will lead to underfitting. Adjusting the value of C helps balance the complexity and generalization ability of the model.
9. The method for predicting hard disk failure in an industrial edge computer room based on IIoT according to claim 1, characterized in that: In step S4, the tuning logic of the gamma parameter is: the larger the gamma value, the smaller the influence range of each data point, and the model becomes more complex and prone to overfitting; the smaller the gamma value, the simpler the model.
10. The method for predicting hard disk failure in an industrial edge computer room based on IIoT according to claim 1, characterized in that: In step S4, the tuning logic of the ε parameter is: when the difference between the predicted value and the actual value is less than ε, the model will not penalize these differences. Conversely, when the difference between the predicted value and the actual value is greater than or equal to ε, the model will penalize these differences. Adjusting the value of ε helps control the sensitivity of the model to errors.
Citation Information
Cited By
State evaluation method and device for solid state disk, storage medium and program product
CN119207540A
