A grain dryer fault detection method and system

Abstract

The application relates to the technical field of intelligent agricultural equipment monitoring, and discloses a grain dryer fault detection method and system. The method comprises the following steps: collecting multi-dimensional time sequence data through a sensor network chip and segmenting the multi-dimensional time sequence data to obtain segmented data sequences; performing frequency domain filtering and distribution analysis on the segmented sequences, separating and removing environmental noise to obtain purified data sequences; performing principal component analysis on the purified data sequences to obtain a main feature set; separating short-term fluctuation signals from the main features, detecting abnormal deviations and generating abnormal fluctuation indexes; when the indexes exceed a warning threshold, the abnormal deviation values are accumulated and calculated, and the average cumulative deviation in a unit window is calculated as an abnormal trend strength; the trend strength is compared with a historical benchmark range, and a fault prediction report is generated and output when the trend strength exceeds the historical benchmark range. Through deep purification, feature extraction and trend quantization analysis of multi-dimensional operation data, the application realizes accurate early warning of early weak faults of a grain dryer.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology for agricultural machinery and equipment, and in particular to a fault detection method and system for grain dryers. Background Technology

[0002] Currently, grain drying is a crucial post-harvest processing step in agricultural production. The stable operation of its equipment directly affects the safety and economic value of grain storage, playing an irreplaceable role in ensuring grain quality and reducing spoilage losses. With the development of agricultural modernization and intelligentization, higher demands are placed on the real-time and precise monitoring of the dryer's operating status.

[0003] In existing technologies, fault detection in grain dryers often relies on single-parameter alarm mechanisms based on fixed thresholds and periodic manual on-site inspections. This method involves deploying single-function temperature or humidity sensors at key locations on the equipment to collect operational data and compare it with preset static safety thresholds, triggering alarms when these thresholds are exceeded. Simultaneously, maintenance personnel conduct on-site inspections at fixed intervals to confirm the equipment's status. This monitoring method, dependent on static rules and human experience, has inherent limitations. Because it lacks the ability to intelligently extract and analyze multi-dimensional time-series data features, it cannot effectively filter out environmental noise to capture weak early signals indicating gradual degradation of equipment performance. It also struggles to perform deep correlation and trend analysis of multiple indicators such as temperature, humidity, and vibration, and cannot meet the demands of fusing and analyzing multi-source heterogeneous time-series data generated by complex monitoring networks composed of sensor chips. Consequently, it lacks deep perception and early warning capabilities regarding the equipment's operating status. As a result, the system has insufficient sensitivity to identify complex fault modes and a high false alarm rate, often only discovering the fault after it has become apparent or caused a shutdown.

[0004] Therefore, the core technical problem facing existing technologies lies in how to leverage the rich data sources provided by modern sensor network chips to perform in-depth cleaning, feature extraction, and trend analysis on multi-dimensional operational data through intelligent algorithms, thereby achieving early and accurate warnings of potential faults in grain dryers and overcoming the shortcomings of traditional detection methods, such as slow response and insensitivity to subtle anomalies. Summary of the Invention

[0005] This invention provides a method and system for detecting faults in grain dryers, which solves the technical problems of existing technologies that rely on fixed thresholds and human experience, resulting in delayed early warning of early and subtle potential faults in grain dryers, and difficulty in accurately identifying abnormal trends in multidimensional and complex data.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a method for detecting faults in a grain dryer, comprising:

[0007] By using a sensor network chip deployed on the grain drying equipment, multi-dimensional time-series data of the grain drying equipment during operation is collected, and the multi-dimensional time-series data is segmented to obtain segmented data sequences.

[0008] Environmental noise and interference signals are separated and removed from the segmented data sequence to obtain a purified data sequence;

[0009] Principal component analysis was performed on the purified data sequence to obtain the main feature set;

[0010] Short-term fluctuation signals are separated from the main feature set, abnormal deviations in the short-term fluctuation signals are detected, and abnormal fluctuation indicators are generated.

[0011] When the abnormal fluctuation index exceeds the preset warning threshold, the abnormal deviation within a continuous time window is cumulatively calculated, and based on the result of the cumulative calculation, the average cumulative deviation within a unit window is calculated as the intensity of the abnormal trend.

[0012] The intensity of the abnormal trend is compared with a predetermined benchmark range. When the intensity of the abnormal trend exceeds the benchmark range, a fault prediction report is generated and output.

[0013] In one optional implementation, the step of collecting multi-dimensional time-series data of the grain drying equipment during operation via a sensor network chip deployed on the equipment, and segmenting the multi-dimensional time-series data to obtain segmented data sequences, includes:

[0014] By deploying sensor network chips on grain drying equipment, temperature data, humidity data and vibration data are collected during the operation of the equipment to obtain multi-dimensional time-series data.

[0015] The multidimensional time-series data is segmented using a sliding window of fixed time length to obtain segmented data sequences.

[0016] In one optional implementation, separating and removing environmental noise and interference signals from the segmented data sequence to obtain a purified data sequence includes:

[0017] The segmented data sequence is converted from the time domain to the frequency domain using Fourier transform, and specific frequency components corresponding to the pre-acquired environmental noise pattern are filtered out. The frequency domain data after noise removal is then subjected to inverse transform to obtain the preliminary filtered signal.

[0018] Data distribution analysis is performed on the preliminary filtered signal to identify and remove residual interference signals whose fluctuation range exceeds a first preset threshold, thereby obtaining the purified data sequence.

[0019] In one optional implementation, the step of detecting abnormal deviations in the short-term fluctuation signal and generating an abnormal fluctuation index includes:

[0020] Calculate the statistical characteristic values ​​of the short-term fluctuation signal within a continuous time window;

[0021] The statistical characteristic values ​​of each time window are compared with the normal fluctuation range of the corresponding window determined based on historical data.

[0022] When the statistical feature value exceeds the corresponding normal fluctuation range, the corresponding window is marked as an abnormal window;

[0023] The abnormal fluctuation index is generated based on the statistical characteristic values ​​of all abnormal windows.

[0024] In one optional implementation, the step of accumulating abnormal deviations within a continuous time window and determining the intensity of the abnormal trend by calculating the average cumulative deviation within a unit window based on the results of the accumulated calculation includes:

[0025] Determine the starting time window when the abnormal fluctuation index first exceeds the preset warning threshold;

[0026] Starting from the initial time window, the abnormal deviation values ​​within N consecutive time windows are weighted and summed to obtain the cumulative deviation value, where N is an integer greater than 1;

[0027] The ratio of the cumulative deviation value to the number of time windows N is calculated as the intensity of the abnormal trend.

[0028] In one optional implementation, generating and outputting the fault prediction report includes:

[0029] Integrate the abnormal pattern information related to the intensity of the abnormal trend, including the abnormal type, occurrence time, and device location;

[0030] A fault prediction report is generated based on the integrated abnormal pattern information, and the fault prediction report is sent to the user terminal.

[0031] Secondly, the present invention provides a grain dryer fault detection system, comprising:

[0032] The data acquisition module is used to collect multi-dimensional time-series data of the grain drying equipment during operation through the sensor network chip deployed on the grain drying equipment, and to segment the multi-dimensional time-series data to obtain segmented data sequences.

[0033] The data purification module is used to separate and remove environmental noise and interference signals from the segmented data sequence to obtain a purified data sequence.

[0034] The feature extraction module is used to perform principal component analysis on the purified data sequence to obtain the main feature set;

[0035] The fluctuation detection module is used to separate short-term fluctuation signals from the main feature set, detect abnormal deviations in the short-term fluctuation signals, and generate abnormal fluctuation indicators.

[0036] The trend quantification module is used to perform cumulative calculation on the abnormal deviation within a continuous time window when the abnormal fluctuation index exceeds a preset warning threshold, and calculate the average cumulative deviation within a unit window as the abnormal trend intensity based on the result of the cumulative calculation.

[0037] The fault early warning module is used to compare the intensity of the abnormal trend with a predetermined benchmark range, and generate and output a fault prediction report when the intensity of the abnormal trend exceeds the benchmark range.

[0038] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the grain dryer fault detection method described in any one of the above.

[0039] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the grain dryer fault detection method described in any one of the above.

[0040] Compared with the prior art, the present invention has the following beneficial effects.

[0041] (1) This invention collects multi-dimensional time-series data through a sensor network chip and performs in-depth purification by combining frequency domain filtering and data distribution analysis, thus constructing a robust data preprocessing mechanism for complex industrial noise environments. The technical derivation of this method lies in the fact that existing methods are difficult to separate real signals from environmental interference. This invention effectively removes multi-source noise through a two-stage noise reduction strategy, thereby solving the problem of weak abnormal signals being submerged by noise and providing a high-quality data foundation for subsequent accurate analysis.

[0042] (2) This invention constructs a progressive fault early warning model based on trend quantization by performing feature dimensionality reduction and short-term fluctuation signal separation on the purified data and calculating the cumulative trend of abnormal deviations. The technical derivation of this method lies in the fact that the existing fixed threshold method cannot capture the performance degradation process of the equipment. This invention quantifies discrete abnormal fluctuations into continuous intensity trends, which can characterize the evolution process of the fault, thereby solving the problem of lagging early warning of potential faults and significantly improving the foresight of fault prediction.

[0043] (3) This invention achieves an automated closed loop from anomaly perception to decision output by dynamically comparing the quantified intensity of abnormal trends with historical benchmarks and automatically generating structured fault reports. The technical derivation of this method lies in the fact that the traditional manual inspection mode is inefficient and highly subjective. By establishing objective quantitative benchmarks and automated reporting processes, this invention reduces the reliance on human experience, thereby supporting centralized and intelligent monitoring of large-scale grain drying equipment and significantly improving operation and maintenance efficiency and reliability. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the grain dryer fault detection method provided in the first embodiment of the present invention;

[0045] Figure 2 This is a flowchart of the core algorithm of the grain dryer fault detection method provided in the first embodiment of the present invention;

[0046] Figure 3 This is a schematic diagram of the structure of the grain dryer fault detection method provided in the second embodiment of the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] Reference Figure 1 The first embodiment of the present invention provides a fault detection method for a grain dryer, comprising the following steps:

[0049] S11, by using the sensor network chip deployed on the grain drying equipment, multi-dimensional time-series data of the grain drying equipment during operation is collected, and the multi-dimensional time-series data is segmented to obtain segmented data sequences;

[0050] S12, Separate and remove environmental noise and interference signals from the segmented data sequence to obtain a purified data sequence;

[0051] S13, perform principal component analysis on the purified data sequence to obtain the main feature set;

[0052] S14, Separate short-term fluctuation signals from the main feature set, detect abnormal deviations in the short-term fluctuation signals, and generate abnormal fluctuation indicators;

[0053] S15, when the abnormal fluctuation index exceeds the preset warning threshold, the abnormal deviation within a continuous time window is cumulatively calculated, and based on the result of the cumulative calculation, the average cumulative deviation within a unit window is calculated as the abnormal trend intensity.

[0054] S16, compare the intensity of the abnormal trend with a predetermined benchmark range, and when the intensity of the abnormal trend exceeds the benchmark range, generate and output a fault prediction report.

[0055] In step S11, multi-dimensional time-series data of the grain drying equipment during operation is collected by a sensor network chip deployed on the grain drying equipment, and the multi-dimensional time-series data is segmented to obtain segmented data sequences, including:

[0056] By deploying sensor network chips on grain drying equipment, temperature data, humidity data and vibration data are collected during the operation of the equipment to obtain multi-dimensional time-series data.

[0057] The multidimensional time-series data is segmented using a sliding window of fixed time length to obtain segmented data sequences.

[0058] First, raw time-series data is collected through sensor network chips deployed at key operating nodes of the grain drying equipment. Temperature sensors are installed at the hot air inlet and outlet of the dryer to collect hot air temperature data, with a measurement range of 0℃ to 200℃ and an accuracy of ±0.2℃. Integrated temperature and humidity sensors are installed inside the grain layer in the drying section to collect internal temperature and humidity data, with a temperature measurement accuracy of ±0.5℃ and a humidity measurement accuracy of ±2%RH. Vibration acceleration sensors are installed in the bearing housing of the drying tower drive motor and the transmission parts of the elevator to collect mechanical vibration data, with a range of ±16g and a frequency response range of 0.5Hz to 1000Hz. The sampling frequency for temperature and humidity data is set to 0.2Hz, based on the physical characteristics of heat and moisture conduction within the grain grains during the drying process. As a porous medium, the change in internal temperature and humidity of grain grains is a relatively slow process, with typical time constants on the order of several minutes to tens of minutes. Setting the sampling interval to 5 seconds effectively captures the key trends of this changing process while avoiding excessively redundant data due to overly rapid sampling, aligning with conventional practices in agricultural engineering for monitoring such slow-changing processes. Those skilled in the art will understand that this frequency can be adjusted within the range of 0.1Hz to 1Hz based on the differences in the thermophysical properties of different grain varieties. The vibration data sampling frequency is set to 1kHz, based on the characteristic frequency range corresponding to typical mechanical equipment faults. For low-speed, heavy-duty bearings and gearboxes commonly used in grain dryers, their fault characteristic frequencies are typically distributed between tens of hertz and below one kilohertz. According to the Nyquist sampling theorem, to reconstruct and analyze signals within this frequency band without distortion, the sampling frequency must be at least twice the target highest frequency. Therefore, setting the sampling frequency to 1kHz ensures coverage and sufficient analysis of the characteristic frequency components of the fault. All sensor data is transmitted via a fieldbus network. Data from each sensor is polled and initially packaged by an embedded data acquisition unit located in the electrical control cabinet, forming a raw data packet containing the device ID, timestamp, sensor type, and measurement value.

[0059] For example, the temperature sensor samples at a frequency of 0.2 Hz, acquiring two temperature data points within the first 5 seconds: 65.3°C and 65.5°C; the corresponding humidity sensor readings are 16.1%RH and 16.0%RH, respectively; the vibration sensor samples at a frequency of 1 kHz, acquiring 5000 acceleration data points (e.g., 0.021g, 0.019g,...) within the same 5 seconds. All data is accompanied by a device ID (e.g., Dryer_001), a timestamp, and a sensor type identifier.

[0060] Subsequently, the acquired multidimensional time-series data is segmented. A sliding window of fixed time length is used to segment the original data packet stream. This method balances the continuity of analysis with computational efficiency. The window length L is set to 300 seconds to capture a statistically significant short-term operating state of the device. The sliding step size S is set to 150 seconds, which is half the window length. For example, for a continuous data stream starting at time T, the first generated window covers the time period [T, T+300 seconds]; the second window covers [T+150 seconds, T+450 seconds]; and so on, with a 150-second overlap between adjacent windows. This overlap ensures continuous and smooth monitoring of the operating state on the time axis, avoiding the omission of brief but important anomalies due to excessively large step sizes, while also effectively controlling the number of generated data segments to avoid excessive computational redundancy due to excessive overlap. The determination of the window length aims to include sufficient data points to calculate stable statistical characteristics, such as mean and variance, while ensuring that the equipment operating state within the window can be considered quasi-steady state, which aligns with the parameter changes of the dryer under stable operating conditions. The determination of the sliding step size is to achieve timely response to new data while ensuring sufficient overlap between adjacent data segments to maintain analytical continuity. This "long window, short step size" configuration is a common strategy in time-series data analysis that balances stability and real-time performance. Those skilled in the art will understand that the specific values ​​of the window and step size can be adjusted within a reasonable range based on the process cycle of different dryer models and the required early warning sensitivity; for example, the window length can be between 2 and 10 minutes, and the step size between 30 seconds and 5 minutes. This operation divides the continuous raw data stream into a series of segmented data sequences with partial temporal overlap. Each sequence contains a time-aligned set of multidimensional data points within a fixed time window, facilitating subsequent short-term state analysis and feature extraction. The segmented data sequences are organized into structured data blocks, serving as the basic input units for subsequent data processing.

[0061] In step S12, environmental noise and interference signals are separated and removed from the segmented data sequence to obtain a purified data sequence, including:

[0062] The segmented data sequence is converted from the time domain to the frequency domain using Fourier transform, and specific frequency components corresponding to the pre-acquired environmental noise pattern are filtered out. The frequency domain data after noise removal is then subjected to inverse transform to obtain the preliminary filtered signal.

[0063] Data distribution analysis is performed on the preliminary filtered signal to identify and remove residual interference signals whose fluctuation range exceeds a first preset threshold, thereby obtaining the purified data sequence.

[0064] First, frequency domain filtering is performed on the segmented data sequences to remove fixed-pattern environmental noise. For each segmented data sequence obtained in step S11, a Fast Fourier Transform algorithm is applied to the temperature, humidity, and vibration data channels to convert the time-domain signal into a frequency-domain spectrum. The pre-acquired environmental noise pattern is established through offline analysis. For example, when the grain drying equipment is in a stopped state or a known stable no-load operation state, a continuous period of sensor background data is collected. Spectral analysis is performed on this background data to extract discrete frequency components that are persistent in all data channels and whose amplitude spectral lines are significantly higher than the floor noise, and these are marked as environmental noise characteristic frequencies. These characteristic frequencies typically include 50Hz power frequency and its harmonic interference generated by other rotating equipment in the workshop, such as fans, as well as power frequency interference introduced by power grid conduction. During online filtering, for the current segmented data sequence to be processed, spectral lines within a preset bandwidth near each of the environmental noise characteristic frequencies are zeroed out or significantly attenuated. The preset bandwidth is exemplarily set to 2Hz. The 2Hz bandwidth setting is based on an estimate of the slight drift or fluctuation that the background noise frequency may exhibit during actual operation, ensuring effective coverage and filtering of this noise component. After filtering all identified noise frequency components, an inverse Fourier transform is performed on the processed spectrum to reconstruct it into a time-domain signal, thus obtaining the preliminary filtered signal. This step effectively removes background noise generated by a fixed source that is unrelated to the operating state of the grain dryer itself.

[0065] Subsequently, data distribution analysis is performed on the preliminary filtered signal to eliminate sudden residual interference. This step aims to identify and remove data segments with abnormal amplitudes caused by instantaneous sensor anomalies, random electromagnetic pulses, etc. For each data channel, the normal fluctuation range of the data points for that channel is determined based on historical normal data. Specifically, the method is as follows: a segment of purified data that is confirmed to be operating without faults is selected from the historical database, and it is divided into multiple segment sequences. For each data channel in each segment sequence, the arithmetic mean of all data points for that channel within that segment sequence is calculated. and standard deviation .in accordance with The criterion is to set the normal fluctuation range of the channel corresponding to the segmented sequence as follows: This range is a two-sided interval used to determine whether a single data point is abnormal. For each segment sequence in the current preliminary filtered signal to be processed, for each data channel, all data points of that channel within this segment sequence are traversed, and those falling within the corresponding normal fluctuation range are counted. The system calculates the number of data points outside this range and their proportion of the total number of data points in that channel within the segmented sequence. For example, suppose a temperature channel segmented sequence has 60 data points, and its normal range calculated based on historical data is [65.0℃, 67.0℃]. After comparison, 18 data points are found to fall outside this range, so the outlier ratio is 18 / 60 = 30%. If this ratio exceeds a first preset threshold, it should be noted that the first preset threshold is set based on the statistical distribution of the outlier ratio in historical normal data. For example, it can be set to the 95th percentile or higher of the statistical value of the outlier ratio in historical normal data to ensure that while filtering out significant interference, it avoids excessive removal of normal fluctuation data. An exemplary threshold range can be between 20% and 40%, in which case it is determined that the segmented sequence has suffered a strong sudden interference in this channel, and the entire data segment of this channel is marked as invalid. After traversing all channels, only those segmented data sequences that have not been marked as invalid in all channels are retained, and they are reassembled in chronological order, finally outputting a purified data sequence for subsequent feature extraction and analysis.

[0066] In step S13, principal component analysis is performed on the purified data sequence to obtain the main feature set.

[0067] First, an initial feature vector is constructed based on the purification data sequence, and dimensionality reduction is performed. For the purification data sequence output in step S12, feature extraction is performed on each segment. Time-domain and frequency-domain statistical features are extracted from the temperature, humidity, and vibration channel data of each segment. Time-domain features include mean, standard deviation, peak-to-peak value, root mean square value, and skewness; frequency-domain features include the mean amplitude spectrum and center frequency within the main energy frequency band after Fast Fourier Transform. For a segmented sequence containing multiple data channels, an initial feature vector is constructed using this method. This vector belongs to the real space. ,in The total number of features is given. Subsequently, principal component analysis is used to reduce the dimensionality of the observation matrix, which consists of the initial eigenvectors of multiple segmented sequences. Specifically, the observation matrix is ​​centered, and then the covariance matrix of the centered data is calculated. By solving the covariance matrix The characteristic equation is used to obtain its eigenvalues. and the corresponding unit eigenvector ,in Sort the eigenvalues ​​in descending order, and denote them as follows: Each feature vector This refers to a principal component direction, where the projection variance of the data along that direction is equal to the corresponding eigenvalue. .

[0068] Subsequently, the principal component contributions are calculated, and key principal components are selected based on a second preset threshold. The contribution of each principal component is then calculated. Contribution of each principal component Defined as the proportion of the variance of this principal component to the total variance of all principal components, its calculation formula is:

[0069]

[0070] In the formula, Indicates the first The contribution of each principal component is a dimensionless scalar. Indicates the first The eigenvalue, i.e. the th eigenvalue. Variance of each principal component direction; Indicates all The sum of all eigenvalues ​​is the total variance of the original data. This represents the total number of dimensions of the initial feature vector.

[0071] forward Cumulative contribution rate of each principal component Then it is the former The sum of contributions is calculated using the following formula:

[0072]

[0073] In the formula, Indicates the preceding The cumulative contribution rate of each principal component is a dimensionless scalar. Indicates the previous Contribution of each principal component arrive Perform summation; The number of principal components selected.

[0074] The second preset threshold A cumulative contribution rate target value is set for screening principal components. The determination of the second preset threshold is based on a balance between model interpretability and computational complexity. For example, [the threshold is set as follows]. The threshold is set at 85%. This threshold is based on the principle of minimizing feature dimensions while ensuring that the dimensionality-reduced data retains most of the variance information of the original data, thereby reducing the complexity of subsequent calculations. During the selection process, the principal components with the largest contributions are accumulated, selecting those that maximize the cumulative contribution rate. The second preset threshold is reached or exceeded for the first time. Minimum number of principal components This Principal Components The constructed projection matrix serves as the basis for the feature transformation, and the set of low-dimensional feature vectors obtained after this projection transformation of the initial feature vectors of all segmented sequences is defined as the main feature set. Each feature vector in this set has a dimension of... Furthermore, the dimensions are independent of each other, collectively representing the most critical information about the equipment's operating status. For example, assuming the calculated contributions of the five principal components are 50%, 20%, 15%, 10%, and 5%, respectively, the cumulative contribution rates are 50%, 70%, 85%, 95%, and 100%. Since when... At that time, cumulative contribution rate First time reaching Therefore, the first three principal components should be selected to form the main feature set.

[0075] In step S14, short-term fluctuation signals are separated from the main feature set, abnormal deviations in the short-term fluctuation signals are detected, and abnormal fluctuation indicators are generated.

[0076] The process of detecting abnormal deviations in the short-term fluctuation signal and generating an abnormal fluctuation index includes:

[0077] Calculate the statistical characteristic values ​​of the short-term fluctuation signal within a continuous time window;

[0078] The statistical characteristic values ​​of each time window are compared with the normal fluctuation range of the corresponding window determined based on historical data.

[0079] When the statistical feature value exceeds the corresponding normal fluctuation range, the corresponding window is marked as an abnormal window;

[0080] The abnormal fluctuation index is generated based on the statistical characteristic values ​​of all abnormal windows.

[0081] First, the statistical characteristic values ​​of the short-term fluctuation signal are calculated, and the normal fluctuation range of each time window is determined. Then, time series smoothing is performed on the main feature set obtained in step S13 to separate its short-term fluctuation signal components. The time series smoothing process exemplarily employs a locally weighted scatter smoothing algorithm. The core of this algorithm is weighted linear regression, with weights set according to explicit quantization rules: for each time point... any point in its neighborhood weight Determined by the Gaussian kernel function relating to time distance, the calculation formula is as follows: ,in This is a bandwidth parameter used to control the rate of weight decay. This rule quantifies the principle of assigning higher weights to neighboring points, ensuring that the smoothing result filters out noise while preserving the true local trend. The difference between the original signal and the smoothed trend component is the short-term fluctuation signal. The short-term fluctuation signal This refers to the residual sequence after removing the long-term trend, which contains instantaneous fluctuation information of the equipment status.

[0082] To quantify the intensity of the fluctuation, it is necessary to calculate the statistical characteristic value of the signal within a continuous time window. For example, the standard deviation of the data within the window is selected as the statistical characteristic value. This method effectively measures the dispersion of data around the mean, providing a direct reflection of fluctuation amplitude. The length of the time window is the same as the segmentation window length in step S11, which is five minutes. By sliding this window, a series of continuous statistical characteristic value sequences can be obtained. ,in This represents the total number of windows.

[0083] The normal fluctuation range for the corresponding window, determined based on historical data, is established using the following method: Short-term fluctuation signals corresponding to the current analysis period are extracted from historical normal operation data. For each specific time window number... Collect a sample set of statistical characteristic values ​​calculated during the window's normal operation over several historical days. ,in This represents the historical number of days. Based on this sample set, its sample mean is calculated. and sample standard deviation According to statistics In principle, the normal fluctuation range of this window Defined as This range is used to determine abnormal deviations from statistical characteristics. For example, for a specific time window of "10:00-10:05 AM", assuming its historical statistical characteristic values... sample mean Sample standard deviation Therefore, the upper limit of its normal fluctuation range is 0.005 + 3 × 0.0015 = 0.0095, and the lower limit is 0.005 - 3 × 0.0015 = 0.0005.

[0084] Subsequently, the feature values ​​are compared with the normal range, abnormal windows are marked, and abnormal fluctuation indicators are generated. For the short-term fluctuation signal to be detected, it is placed within a time window. Statistical characteristic values ​​obtained by internal calculation This is consistent with the normal fluctuation range predetermined by the window. A comparison is performed. If the current feature value exceeds the normal fluctuation range, the window is determined to have an abnormal deviation and is marked as an abnormal window. All marked abnormal windows are collected to form an abnormal window set A. For example, if the current day's value is calculated within the aforementioned "10:00-10:05" window... Since 0.011 > 0.0095, the window is determined to be an abnormal window.

[0085] Based on the abnormal window set Generate the abnormal fluctuation index For example, the abnormal fluctuation indicator It can be defined as the period after the most recent fixed observation. Within the window, the proportion of abnormal windows to the total number of windows is calculated using the following formula:

[0086]

[0087] in, Indicates the duration of observation The number of abnormal windows within, express The total number of time windows included. This metric It quantifies the intensity of abnormal fluctuations in the short term; the higher the value, the more significantly the equipment's operating status deviates from the normal mode.

[0088] In step S15, when the abnormal fluctuation index exceeds the preset warning threshold, the abnormal deviation within a continuous time window is cumulatively calculated, and based on the result of the cumulative calculation, the intensity of the abnormal trend is determined by calculating the average cumulative deviation within a unit window.

[0089] This includes accumulating abnormal deviations within a continuous time window, and determining the intensity of the abnormal trend by calculating the average cumulative deviation within a unit window based on the results of the accumulated calculation, including:

[0090] Determine the starting time window when the abnormal fluctuation index first exceeds the preset warning threshold;

[0091] Starting from the initial time window, the abnormal deviation values ​​within N consecutive time windows are weighted and summed to obtain the cumulative deviation value, where N is an integer greater than 1;

[0092] The ratio of the cumulative deviation value to the number of time windows N is calculated as the intensity of the abnormal trend.

[0093] First, when the abnormal fluctuation index exceeds a preset warning threshold, a starting window for cumulative calculation is determined, and the abnormal deviation value for each window is calculated. The preset warning threshold... The setting is not a fixed value, but rather based on statistical analysis of historical operating data and a balance between false alarm rate and false negative rate; for example, the value is calculated by the data collection equipment over a large number of normal operating cycles. Historical sequences, and the data corresponding to the hours preceding a recordable equipment failure. Sequence. By analyzing the distribution of normal sequences and comparing the evolution characteristics of pre-fault sequences, a critical value that can best distinguish normal fluctuations from early fault characteristics is selected as... If the 99th percentile of a normal sequence is 0.25, and most failures occur before... If it continues to exceed 0.3, then it can be... A value is set between 0.25 and 0.3, such as 0.28. For example, the abnormal fluctuation indicator... First time exceeding the warning threshold The five-minute time window corresponding to the given time point is determined as the starting time window. Starting from this starting window, N consecutive time windows are selected for subsequent calculations. For example, N is set to 6, corresponding to a 30-minute observation period. This duration is considered effective in capturing an early, continuously developing anomaly while maintaining a sensitive response to changes in state.

[0094] In response to this Each window in the window Calculate its abnormal deviation value The abnormal deviation value Statistical characteristic values ​​used to quantify this window The degree to which it deviates from its normal fluctuation range. Let the lower limit of the normal fluctuation range of this window be... The upper limit is Abnormal deviation value in accordance with and interval The relative position is determined by the following formula:

[0095] like ,but ;

[0096] like ,but ;

[0097] like ,but .

[0098] thus, It is a non-negative value. When Below the lower limit of the normal range hour, It equals the difference between its lower limit and the lower limit, representing the degree of negative deviation; when When within the normal range, A value of zero indicates no significant deviation; when Higher than the upper limit of the normal range hour, The difference between the value and the upper limit represents the degree of positive deviation. This method calculates the unidirectional distance from the feature value to the boundary of the normal range, covering both abnormally low and abnormally high cases, ensuring that subsequent cumulative calculations can comprehensively reflect the severity of various abnormal deviations. For example, suppose the upper limit of the normal range for a certain window... Current eigenvalue ,but .like If it is within the normal range, then .like Below the lower limit ,but .

[0099] Subsequently, the outlier deviations within the continuous time window are weighted and summed to obtain the cumulative deviation value. The purpose of weighted summation is to assign higher weight to recent outlier deviations during the accumulation process, so as to more sensitively reflect the latest changes in the trend. The formula for calculating the cumulative deviation value is:

[0100]

[0101] in, This represents the cumulative deviation value. This represents the total number of consecutive time windows participating in the accumulation. Indicates the first The weighting coefficients of each window enable the accumulation process to reflect abnormal development trends rather than simple arithmetic sums. Indicates the first Abnormal deviation values ​​for each window. Weighting coefficients. The setting exemplifies a linear increment, that is... This means that for the sequence number is The window, When the value is 1, it is the starting window. When the value equals N, it is the nearest window, and its weights are from... The linear increase to 1 assigns a higher weight to recent anomalous deviations, thus increasing the cumulative deviation value. It depends not only on the total amount of the abnormal deviation, but also more sensitively reflects whether the abnormality has recently shown an intensifying trend. If the abnormality has recently intensified, that is... exist Larger time values ​​result in higher values, and even if the total deviation is not large, it will still produce higher values. Value. For example, let's say... Abnormal deviation values ​​of the four windows Given [0, 0.002, 0.001, 0.004], the corresponding weights are... They are [1 / 4, 2 / 4, 3 / 4, 4 / 4], respectively. Cumulative deviation value.

[0102] Finally, the average cumulative deviation within a unit window is calculated as the anomaly trend strength. Anomaly trend strength quantifies the degree of anomalous deviation exhibited on average for each time window, and is calculated using the following formula:

[0103]

[0104] in, Indicates the strength of the abnormal trend; is a scalar value. This represents the cumulative deviation value obtained by weighted summation; This indicates the number of time windows involved in the calculation. This metric... It comprehensively reflects the overall level of abnormal deviation and its concentration and growth trend over time. The larger the value, the more rapidly the equipment status is deviating from the normal mode, and the higher the risk of failure.

[0105] In step S16, the intensity of the abnormal trend is compared with a predetermined benchmark range, and when the intensity of the abnormal trend exceeds the benchmark range, a fault prediction report is generated and output.

[0106] The process of generating and outputting a fault prediction report includes:

[0107] Integrate the abnormal pattern information related to the intensity of the abnormal trend, including the abnormal type, occurrence time, and device location;

[0108] A fault prediction report is generated based on the integrated abnormal pattern information, and the fault prediction report is sent to the user terminal.

[0109] First, the intensity of the abnormal trend is compared with a predetermined benchmark range. The intensity of the abnormal trend from step S15 is then received. .Will The system compares the data with a predetermined baseline range to determine whether a final warning is triggered. This baseline range is determined based on historical data statistics of abnormal trend intensity calculated under long-term (at least one complete production cycle) healthy operating conditions. For example, all calculated data during fault-free operation of the equipment are collected. Value samples, calculate the mean of the sample set. and standard deviation Set the upper limit of the baseline range to The lower limit is set to 0. This range, based on statistical principles, covers the vast majority of normal fluctuations. The value, whose range is from 0 to When the conditions are met When the intensity of the abnormal trend is determined to be significantly beyond the normal baseline range, the equipment has a clear potential risk of failure, and a failure prediction report must be generated immediately.

[0110] Subsequently, multi-dimensional anomaly pattern information related to this determination is integrated. This integration step is executed automatically after the determination is triggered. The integrated anomaly pattern information constitutes the core data foundation for fault diagnosis and localization. The specific integration content includes three dimensions: the first is the anomaly type, whose determination logic is logically linked to the dominant anomaly fluctuation data channel identified in step S14. The system analyzes the anomaly deviation value of each sensor data channel within the main contributing window constituting the current high trend intensity. The sum of the abnormal deviation values ​​of each channel is calculated as a percentage of the total abnormal deviation values ​​of all channels. If the percentage of a certain channel exceeds a preset threshold, such as 60%, then that channel is determined to be the dominant channel and marked as an abnormal trend of the corresponding type. If the temperature channel is dominant, it is marked as a temperature abnormal trend; if the vibration channel is dominant, it is marked as a vibration abnormal trend. If the percentage of any channel does not exceed the threshold, then it is determined that multiple channels are jointly dominant and marked as a composite abnormal trend. The second is the timeline of occurrence. The system accurately records two key timestamps: one is the current system time when this determination takes effect, and the other is the time window from which the abnormal trend started, as determined in step S15. This allows for a complete depiction of the duration from the appearance of an anomaly to its development into a warning level. The third part is the device location, which is obtained and recorded by associating it with the unique identifier of the device corresponding to the data stream that triggered this judgment.

[0111] Finally, a structured fault prediction report is generated and sent. The integrated anomaly pattern information is automatically filled into a preset structured report template, generating a complete report including a report title, generation time, equipment identifier, equipment location, anomaly type, anomaly trend intensity value, trend start time, risk level assessment, and preliminary maintenance recommendations. The risk level assessment is based on the anomaly trend intensity... The degree to which the value exceeds the upper limit of the baseline range is divided into three levels: Attention, Warning, and Severe. For example, a risk coefficient is defined. .like It is rated as a level of attention; if It is rated as a warning level; if If an anomaly is detected, it is classified as a severe level. This classification method measures the current anomaly within the context of historical fluctuations, making risk assessment more objective and comparable. Other fields in the report, such as anomaly type, time, location, trend strength value, and start time, are directly filled in with the corresponding information. After the report is generated, it is automatically and in real time to pre-configured receiving terminals through the system-integrated communication interface, according to preset push rules, in the form of structured data messages or formatted documents. For example, reports at the attention level can be pushed to the mobile application of on-duty personnel; reports at the warning and severe levels are simultaneously pushed to the mobile application, the central monitoring screen, and the email of the operations manager, ensuring that key information is received in a timely and reliable manner. This completes a full intelligent early warning closed loop from anomaly detection, trend analysis, risk assessment to information delivery.

[0112] Reference Figure 2 The first embodiment of the present invention provides a flowchart of the core algorithm for implementing the method of the present invention. This schematic diagram fully presents the key quantitative process from the triggering of abnormal fluctuations to the calculation of the intensity of abnormal trends.

[0113] The process first determines whether the abnormal fluctuation index output in step S14 exceeds a preset warning threshold. The warning threshold is determined by the statistical distribution of historical normal operation data and pre-fault abnormal patterns, effectively distinguishing normal fluctuations from precursory signals of potential faults. When the abnormal fluctuation index first exceeds this threshold, the system records the corresponding time window and uses it as the starting point for cumulative deviation analysis, laying the data benchmark for subsequent trend quantification.

[0114] Subsequently, the process enters the cumulative calculation stage of abnormal deviations. The algorithm starts with the initial time window and selects N consecutive time windows to form a fixed-length observation interval. Within this interval, the system extracts the statistical characteristic values ​​of each time window and calculates its abnormal deviation based on the upper limit of the normal fluctuation range. By applying linearly increasing weights to the deviations of each window and summing them, a cumulative deviation value that reflects the abnormal development trajectory is generated, making the contribution of recent deviations to the overall trend more significant, thereby enhancing the sensitivity and real-time performance of trend quantification.

[0115] After obtaining the cumulative deviation, the system calculates the average cumulative deviation per unit window to generate the final abnormal trend strength. This trend strength index comprehensively reflects the scale, duration, and growth trend of the abnormal deviation, and is an important basis for determining whether the equipment has entered the failure development stage. A larger trend strength means that the equipment deviates more from the normal operating mode, and the failure risk is accumulating at an accelerated pace.

[0116] The entire process constitutes the core algorithm chain of "threshold triggering → deviation accumulation → trend quantification", which provides the basic support for the invention to accurately characterize early abnormal trends and provides key input data for subsequent fault prediction and early warning strategies.

[0117] Reference Figure 3 The second embodiment of the present invention provides a grain dryer fault detection system, comprising:

[0118] The data acquisition module is used to collect multi-dimensional time-series data of the grain drying equipment during operation through the sensor network chip deployed on the grain drying equipment, and to segment the multi-dimensional time-series data to obtain segmented data sequences.

[0119] The data purification module is used to separate and remove environmental noise and interference signals from the segmented data sequence to obtain a purified data sequence.

[0120] The feature extraction module is used to perform principal component analysis on the purified data sequence to obtain the main feature set;

[0121] The fluctuation detection module is used to separate short-term fluctuation signals from the main feature set, detect abnormal deviations in the short-term fluctuation signals, and generate abnormal fluctuation indicators.

[0122] The trend quantification module is used to perform cumulative calculation on the abnormal deviation within a continuous time window when the abnormal fluctuation index exceeds a preset warning threshold, and calculate the average cumulative deviation within a unit window as the abnormal trend intensity based on the result of the cumulative calculation.

[0123] The fault early warning module is used to compare the intensity of the abnormal trend with a predetermined benchmark range, and generate and output a fault prediction report when the intensity of the abnormal trend exceeds the benchmark range.

[0124] It should be noted that the grain dryer fault detection system provided in this embodiment of the invention is used to execute all the process steps of the grain dryer fault detection method in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0125] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps described in the method embodiments above.

[0126] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0127] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0128] The processor 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 any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0129] The memory can be used to store the computer programs and modules. The processor implements various functions of the electronic device by running or executing the computer programs and modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0130] If the modules / units integrated into the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or system capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0131] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0132] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for detecting faults in a grain dryer, characterized in that, include: By using a sensor network chip deployed on the grain drying equipment, multi-dimensional time-series data of the grain drying equipment during operation is collected, and the multi-dimensional time-series data is segmented to obtain segmented data sequences. Environmental noise and interference signals are separated and removed from the segmented data sequence to obtain a purified data sequence; Principal component analysis was performed on the purified data sequence to obtain the main feature set; Short-term fluctuation signals are separated from the main feature set, abnormal deviations in the short-term fluctuation signals are detected, and abnormal fluctuation indicators are generated. When the abnormal fluctuation index exceeds the preset warning threshold, the abnormal deviation within a continuous time window is cumulatively calculated, and based on the result of the cumulative calculation, the average cumulative deviation within a unit window is calculated as the intensity of the abnormal trend. The intensity of the abnormal trend is compared with a predetermined benchmark range. When the intensity of the abnormal trend exceeds the benchmark range, a fault prediction report is generated and output. The step of accumulating the abnormal deviations within a continuous time window, and calculating the average cumulative deviation within a unit window as the intensity of the abnormal trend based on the results of the cumulative calculation, includes: Determine the starting time window when the abnormal fluctuation index first exceeds the preset warning threshold; Starting from the initial time window, the abnormal deviation values ​​within N consecutive time windows are weighted and summed to obtain the cumulative deviation value, where N is an integer greater than 1; The ratio of the cumulative deviation value to the number of time windows N is calculated as the intensity of the abnormal trend; The process of detecting abnormal deviations in the short-term fluctuation signal and generating an abnormal fluctuation index includes: Calculate the statistical characteristic values ​​of the short-term fluctuation signal within a continuous time window; The statistical characteristic values ​​of each time window are compared with the normal fluctuation range of the corresponding window determined based on historical data. When the statistical feature value exceeds the corresponding normal fluctuation range, the corresponding window is marked as an abnormal window; The abnormal fluctuation index is generated based on the statistical feature values ​​of all abnormal windows. The abnormal fluctuation index is the proportion of the number of abnormal windows to the total number of windows within the most recent fixed observation period.

2. The grain dryer fault detection method according to claim 1, characterized in that, The process involves collecting multi-dimensional time-series data of the grain drying equipment during operation via a sensor network chip deployed on the equipment, and segmenting the multi-dimensional time-series data to obtain segmented data sequences, including: By deploying sensor network chips on grain drying equipment, temperature data, humidity data and vibration data during equipment operation are collected to obtain multi-dimensional time-series data. The multidimensional time-series data is segmented using a sliding window of fixed time length to obtain segmented data sequences.

3. The grain dryer fault detection method according to claim 1, characterized in that, The step of separating and removing environmental noise and interference signals from the segmented data sequence to obtain a purified data sequence includes: The segmented data sequence is converted from the time domain to the frequency domain using Fourier transform, and specific frequency components corresponding to the pre-acquired environmental noise pattern are filtered out. The frequency domain data after noise removal is then subjected to inverse transform to obtain the preliminary filtered signal. Data distribution analysis is performed on the preliminary filtered signal to identify and remove residual interference signals whose fluctuation range exceeds a first preset threshold, thereby obtaining the purified data sequence.

4. The grain dryer fault detection method according to claim 1, characterized in that, The generation and output of the fault prediction report includes: Integrate the abnormal pattern information related to the intensity of the abnormal trend, including the abnormal type, occurrence time, and device location; A fault prediction report is generated based on the integrated abnormal pattern information, and the fault prediction report is sent to the user terminal.

5. A fault detection system for a grain dryer, characterized in that, A method for detecting faults in a grain dryer as described in any one of claims 1 to 4, comprising: The data acquisition module is used to collect multi-dimensional time-series data of the grain drying equipment during operation through the sensor network chip deployed on the grain drying equipment, and to segment the multi-dimensional time-series data to obtain segmented data sequences. The data purification module is used to separate and remove environmental noise and interference signals from the segmented data sequence to obtain a purified data sequence. The feature extraction module is used to perform principal component analysis on the purified data sequence to obtain the main feature set; The fluctuation detection module is used to separate short-term fluctuation signals from the main feature set, detect abnormal deviations in the short-term fluctuation signals, and generate abnormal fluctuation indicators. The trend quantification module is used to perform cumulative calculation on the abnormal deviation within a continuous time window when the abnormal fluctuation index exceeds a preset warning threshold, and calculate the average cumulative deviation within a unit window as the abnormal trend intensity based on the result of the cumulative calculation. The fault early warning module is used to compare the intensity of the abnormal trend with a predetermined benchmark range, and generate and output a fault prediction report when the intensity of the abnormal trend exceeds the benchmark range.

Images