Sensor network-based carton production equipment state predictive maintenance method

Abstract

The application discloses a carton production equipment state predictive maintenance method based on a sensor network and belongs to the technical field of equipment monitoring. Vibration signals are collected at key positions of the equipment and vibration time series are generated; vibration amplitudes in each window are extracted to construct a state feature vector and form a state feature sequence; current operation condition parameters of the equipment are acquired; historical operation records of the same condition are screened; absolute value accumulations of component difference values of corresponding positions of each historical state feature sequence are calculated to obtain difference values, and the smallest historical record is selected as a matching record; the remaining normal operation time of the current equipment is predicted according to a fault occurrence time, and a maintenance prompt instruction is generated when the remaining normal operation time is lower than a threshold value. The application significantly improves the fault early warning accuracy under different production speeds and raw material thickness conditions, avoids false alarms of a static threshold value, effectively suppresses false triggering caused by instantaneous abnormal fluctuations, realizes progressive self-optimization of prediction accuracy, and is suitable for intelligent predictive maintenance of a carton continuous production line.

Description

Technical Field

[0001] This invention discloses a predictive maintenance method for the condition of cardboard box production equipment based on sensor networks, belonging to the field of equipment monitoring technology. Background Technology

[0002] Cardboard box production equipment, such as corrugated board production lines, crimping machines, die-cutting machines, and roller laminating machines, are core processing equipment in the packaging industry, and their operating status directly affects product quality and production efficiency. Currently, the maintenance and management of this type of equipment mainly adopts two traditional models: one is regular preventive maintenance, which involves shutting down the equipment for inspection and replacement of parts according to fixed time cycles, such as monthly or quarterly; the other is reactive fault repair, which only arranges maintenance personnel to handle the equipment after obvious abnormalities or complete shutdowns. However, regular maintenance often ignores the differences in actual operating conditions of the equipment. During periods of low production load, this may lead to over-maintenance and waste of resources, while during peak production periods, excessively long maintenance cycles may result in faults not being addressed in a timely manner. Reactive repair directly leads to unplanned downtime, causing serious consequences such as raw material scrapping and delivery delays. Therefore, there is an urgent need for a technical solution that can dynamically assess the health status of the equipment based on its real-time operating status and provide early maintenance warnings.

[0003] With the development of sensor technology and the Industrial Internet of Things, some advanced manufacturing companies have begun to install vibration sensors on key parts of cardboard box production equipment, such as bearing housings, transmission gearboxes, and roller support ends. By collecting vibration signals and performing spectrum analysis or time-domain statistics, they can identify abnormal vibration patterns.

[0004] However, most existing condition monitoring methods are limited to simple threshold alarms, which trigger an alarm when the vibration amplitude exceeds a fixed upper limit. This static threshold approach cannot adapt to normal vibration fluctuations of equipment under different production speeds and raw material thicknesses, easily leading to a large number of false alarms or missed alarms. In addition, some improved solutions introduce machine learning or deep neural networks for fault prediction, but these methods require a large number of labeled fault samples and long-term model training, and model updates rely on offline retraining, making it difficult to quickly migrate to new operating conditions or new equipment models. The implementation cost is high, making it difficult for small and medium-sized cardboard packaging companies to deploy practically. Summary of the Invention

[0005] While existing maintenance approaches based on similarity matching of historical fault records are theoretically feasible, they still face numerous unresolved issues in practical applications. For example, vibration characteristics differ significantly under different operating conditions; indiscriminately using all historical data for matching severely reduces accuracy. Furthermore, the length of vibration sequences acquired in real-time often differs from those in historical records, and existing methods lack effective length alignment mechanisms. Additionally, most solutions only consider differences in vibration amplitude, neglecting the crucial impact of vibration trend direction on fault type identification, and failing to assign appropriate weights to different statistical features such as maximum, minimum, and average values. More critically, existing methods lack dynamic confirmation after outputting maintenance prompts, making it difficult to prevent erroneous maintenance commands triggered by instantaneous abnormal fluctuations. Moreover, the historical database cannot self-update after a fault occurs, hindering sustained improvement in prediction accuracy over long-term use. Therefore, there is an urgent need in this field for a predictive maintenance method that comprehensively considers operating condition selection, adaptive sequence length alignment, trend direction discrimination, and weighted difference assessment to overcome these shortcomings.

[0006] To achieve the above objectives, this application provides the following technical solution:

[0007] A method for predictive maintenance of carton and box production equipment based on sensor networks, comprising:

[0008] S1, by continuously collecting vibration signals during the operation of the equipment through sensor nodes deployed in key parts of the carton and paper box production equipment, the vibration amplitude values ​​acquired at each sampling moment are arranged in the order of collection time to generate a vibration time series;

[0009] S2, the vibration time series is divided into multiple consecutive time windows according to a preset time window length, the maximum value, minimum value and average value of vibration amplitude are extracted in each time window, and the maximum value, minimum value and average value in the same time window are combined into the state feature vector corresponding to the time window to form a state feature sequence;

[0010] S3, obtain the current production speed parameters and raw material thickness parameters of the cardboard box production equipment, use the production speed parameters and raw material thickness parameters as working condition identifiers, filter out historical operating records that are the same as the current working condition identifiers from the historical database, each historical operating record contains a historical state feature sequence and the corresponding fault occurrence time node and fault type;

[0011] S4, calculate the difference between the corresponding elements of the current state feature sequence and the historical state feature sequence in each historical running record by summing the differences between the current state feature sequence and each historical state feature sequence, and select the historical running record corresponding to the historical state feature sequence with the smallest difference as the matching record;

[0012] S5. Based on the fault occurrence time node and the current time node in the matching record, calculate the time interval between the fault occurrence time node and the current time node, and use this time interval as the predicted remaining normal operation time. When the predicted remaining normal operation time is lower than the preset threshold, generate a maintenance prompt instruction and output it to the device management terminal.

[0013] Furthermore, the time window length in S2 is dynamically determined based on the periodic pulses in the vibration signal collected by the sensor node, including:

[0014] The pulse moments in the vibration time series where the amplitude of two adjacent peak values ​​exceeds the amplitude threshold are detected, the time interval between adjacent pulse moments is calculated as the equipment rotation period, and the length of the time window is set as an integer multiple of the rotation period.

[0015] Furthermore, the difference accumulation calculation in S4 specifically involves:

[0016] Calculate the absolute values ​​of the differences between the current state feature vector and the corresponding state feature vector in the historical state feature sequence, including the absolute values ​​of the maximum, minimum, and average component differences. Then, add the three absolute values ​​of the differences at the same position to obtain the position difference value at that position. Finally, sum the position difference values ​​of all positions to obtain the total difference value.

[0017] Furthermore, in step S4, when the length of the current state feature sequence is inconsistent with the length of the historical state feature sequence, the following length alignment operation is performed:

[0018] If the length of the current state feature sequence is less than the length of the historical state feature sequence, then a prefix subsequence with the same length as the current state feature sequence is extracted from the beginning of the historical state feature sequence, and this prefix subsequence is used as the historical state feature sequence to participate in the difference accumulation calculation.

[0019] If the length of the current state feature sequence is greater than the length of the historical state feature sequence, then the starting end of the current state feature sequence is used as the reference, and a prefix subsequence with the same length as the historical state feature sequence is extracted from the current state feature sequence as the current state feature sequence to participate in the difference accumulation calculation.

[0020] After length alignment is completed, the maximum value component, minimum value component, and average value component of each state feature vector in the aligned current state feature sequence are taken in chronological order. The absolute value of the difference is calculated with the maximum value component, minimum value component, and average value component of the corresponding position state feature vector in the aligned historical state feature sequence. The absolute values ​​of the differences of the three components at the same position are added together to obtain the difference component at that position. The difference components at all positions are then summed to obtain the total difference value.

[0021] For each historical running record in the historical database, the above length alignment and total difference value calculation are performed to obtain the total difference value between the current state feature sequence and each historical state feature sequence. Then, all total difference values ​​are sorted, and the historical running record corresponding to the historical state feature sequence with the smallest total difference value after sorting is determined as the matching record.

[0022] Furthermore, the preset threshold in S5 is the average running time calculated based on the fault occurrence time node and equipment start running time node corresponding to all historical operation records with the same identifier as the current operating condition in the historical database. The average running time is multiplied by a preset proportional coefficient to obtain the value as the preset threshold.

[0023] Furthermore, each historical operation record in the historical database in S3 also includes the total cumulative operating time of the equipment corresponding to the record. After filtering out historical operation records with the same identifier as the current operating condition from the historical database, the total cumulative operating time of the current equipment is obtained, and the absolute value of the difference between the total cumulative operating time of the current equipment and the total cumulative operating time of the equipment in each filtered historical operation record is calculated. A preset number of historical operation records with the smallest absolute value of the difference are selected from the filtered historical operation records, and the preset number of historical operation records are used as the objects for difference accumulation calculation in S4.

[0024] Furthermore, the specific process of generating maintenance prompt instructions in S5 is as follows:

[0025] The fault type in the matching record is obtained as the expected fault type. At the same time, the fault occurrence time node in the matching record is obtained. The difference between the fault occurrence time node and the current time node is calculated to obtain the predicted remaining normal operation time.

[0026] The predicted remaining normal operating time is compared with multiple pre-divided time intervals to determine the time interval to which the predicted remaining normal operating time belongs. Each time interval corresponds to a pre-set maintenance urgency level. Based on the determined maintenance urgency level, the corresponding maintenance operation description text is extracted from the maintenance measures table. The maintenance measures table stores the correspondence between different maintenance urgency levels and maintenance operation description texts.

[0027] The expected fault type, the maintenance urgency level, and the maintenance operation description text are combined into structured maintenance prompt data;

[0028] After generating maintenance prompts and outputting them to the equipment management terminal, the vibration time sequence at the current moment, the current state characteristic sequence, the identifier of the matching record, and the current time node are also stored as new historical operation records in the historical database.

[0029] The new historical operation record marks the status of the device as not yet having failed. When the device actually fails later, the actual failure time and the actual failure type will be updated to the historical operation record.

[0030] Furthermore, in the difference accumulation calculation process in S4, different weighting coefficients are assigned to the maximum value component, the minimum value component, and the average value component, specifically as follows:

[0031] Obtain the production speed parameters in the current S3, and determine the weight configuration based on the speed range in which the production speed parameters are located. The weight configuration specifies the weight values ​​of the maximum value component, the minimum value component, and the average value component.

[0032] When calculating the difference components at each location, the absolute value of the difference of the maximum value component is multiplied by the weight of the maximum value component to obtain the weighted maximum value difference; the absolute value of the difference of the minimum value component is multiplied by the weight of the minimum value component to obtain the weighted minimum value difference; the absolute value of the difference of the average value component is multiplied by the weight of the average value component to obtain the weighted average value difference; then the weighted maximum value difference, weighted minimum value difference, and weighted average value difference are added together to obtain the difference component at that location; finally, the difference components at all locations are summed to obtain the difference value.

[0033] Furthermore, in step S5, after generating the maintenance prompt command and outputting it to the device management terminal, maintenance confirmation is also performed:

[0034] During the preset waiting period after the output maintenance prompt command, vibration signals continue to be collected through the sensor node, and the difference between the current state feature sequence and each historical state feature sequence in the historical database is recalculated according to the methods of S2 to S4. Matching records are reselected, and the remaining normal operation time is recalculated based on the reselected matching records.

[0035] If the recalculated predicted remaining normal operating time is still lower than the preset threshold, the previously output maintenance prompt instruction remains unchanged; if the recalculated predicted remaining normal operating time is not lower than the preset threshold, a cancellation instruction is sent to the device management terminal to cancel the previously output maintenance prompt instruction.

[0036] Furthermore, in S3, the historical operation records in the historical database are divided into multiple fault category sub-databases according to fault type, and each fault category sub-database stores a sequence of historical state features with the same fault type;

[0037] Before selecting the historical state feature sequence with the smallest difference in S4, the trend of change of the maximum value component and the average value component in the current state feature sequence is determined. The trend of change is determined as follows:

[0038] Take the state feature vectors of the last three time windows in the current state feature sequence, and calculate the successive difference signs of the maximum value component and the average value component in these three time windows respectively. If the successive difference signs of the maximum value component and the average value component are both positive, the trend direction is determined to be an upward trend; if both are negative, it is determined to be a downward trend; otherwise, it is determined to be a stable trend. After obtaining the trend direction of the current state feature sequence, candidate records with the same trend direction as its historical state feature sequence are selected from the historical database. The trend direction of the historical state feature sequence is pre-stored in the historical database. Then, the difference accumulation calculation and minimum difference selection of S4 are performed on the candidate records.

[0039] This invention discloses a predictive maintenance method for cardboard box and carton production equipment based on sensor networks, belonging to the field of equipment monitoring technology. Vibration signals are collected from key parts of the equipment to generate vibration time series; vibration amplitude within each window is extracted to construct a state feature vector, forming a state feature sequence; current operating parameters of the equipment are obtained; historical operating records with the same operating conditions are filtered; the absolute values ​​of the differences between the corresponding positional components of each historical state feature sequence and the observed values ​​are accumulated to calculate the difference value, and the historical record with the smallest difference is selected as the matching record; the remaining normal operating time of the current equipment is predicted based on the fault occurrence time, and a maintenance prompt instruction is generated when it falls below a threshold. This invention significantly improves the accuracy of fault warnings under different production speeds and raw material thicknesses, avoids false alarms caused by static thresholds, effectively suppresses false triggers caused by instantaneous abnormal fluctuations, and achieves progressive self-optimization of prediction accuracy, making it suitable for intelligent predictive maintenance of continuous cardboard box and carton production lines. Attached Figure Description

[0040] Figure 1 A flowchart illustrating the process of a sensor network-based predictive maintenance method for cardboard box production equipment, as claimed in an embodiment of the present invention.

[0041] Figure 2 The second workflow diagram is shown for a sensor network-based predictive maintenance method for carton and paper box production equipment, as claimed in an embodiment of the present invention.

[0042] Figure 3The third flowchart is a method for predictive maintenance of carton and box production equipment based on sensor networks, as claimed in an embodiment of the present invention. Detailed Implementation

[0043] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0044] The terms "first," "second," and "third" in this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of those features. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications in the embodiments of this application, such as up, down, left, right, front, back, etc., are only used to explain the relative positional relationships and movements between components in a specific orientation as shown in the accompanying drawings. If the specific orientation changes, the directional indications will change accordingly. Furthermore, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0045] References to embodiments herein mean that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0046] According to the first embodiment of the present invention, referring to Figure 1 This invention claims protection for a predictive maintenance method for the condition of cardboard box and carton production equipment based on sensor networks, comprising:

[0047] S1, by continuously collecting vibration signals during the operation of the equipment through sensor nodes deployed in key parts of the carton and paper box production equipment, the vibration amplitude values ​​acquired at each sampling moment are arranged in the order of collection time to generate a vibration time series;

[0048] S2, the vibration time series is divided into multiple consecutive time windows according to a preset time window length, the maximum value, minimum value and average value of vibration amplitude are extracted in each time window, and the maximum value, minimum value and average value in the same time window are combined into the state feature vector corresponding to the time window to form a state feature sequence;

[0049] S3, obtain the current production speed parameters and raw material thickness parameters of the cardboard box production equipment, use the production speed parameters and raw material thickness parameters as working condition identifiers, filter out historical operating records that are the same as the current working condition identifiers from the historical database, each historical operating record contains a historical state feature sequence and the corresponding fault occurrence time node and fault type;

[0050] S4, calculate the difference between the corresponding elements of the current state feature sequence and the historical state feature sequence in each historical running record by summing the differences between the current state feature sequence and each historical state feature sequence, and select the historical running record corresponding to the historical state feature sequence with the smallest difference as the matching record;

[0051] S5. Based on the fault occurrence time node and the current time node in the matching record, calculate the time interval between the fault occurrence time node and the current time node, and use this time interval as the predicted remaining normal operation time. When the predicted remaining normal operation time is lower than the preset threshold, generate a maintenance prompt instruction and output it to the device management terminal.

[0052] In this embodiment, a high-speed corrugated cardboard crimping machine in a cardboard box production workshop is designated as the target equipment.

[0053] A piezoelectric accelerometer is installed on the surface of the spindle front bearing housing of the device. The sensor's data acquisition card continuously collects vibration signals from the device during its stable operation phase after no-load startup at a sampling frequency of 2048 times per second. The acquisition duration is ten seconds, thus obtaining a raw vibration time series containing 20480 discrete vibration amplitude data points. These amplitude data points are in microvolts and are arranged in chronological order from the zeroth millisecond to the ten-thousandth millisecond. After S1 is completed, S2 proceeds, dividing the entire ten-second vibration time series into 20 consecutive and non-overlapping time windows of 500 milliseconds each.

[0054] For the first time window, from 0 milliseconds to 500 milliseconds, the maximum value (let's say 15.3 microvolts) and the minimum value (let's say 11.7 microvolts) are found from the 1024 amplitude data points contained within this interval. The arithmetic mean of all 1024 amplitude data points in this interval is then calculated (let's say 13.5 microvolts). These three values ​​are combined in a fixed order of maximum, minimum, and average to form the first state feature vector corresponding to this time window. The same maximum, minimum, and average extraction operation is performed sequentially on the remaining 19 time windows to obtain the second state feature vector up to the 20th state feature vector. These vectors are arranged in the chronological order of the time windows, forming a state feature sequence of the current device operating state.

[0055] In S3, the host computer of the equipment management system obtains the production speed parameter displayed on the current crimping machine's operating panel. This parameter is 192 finished cardboard boxes per minute. Simultaneously, it obtains the raw material cardboard thickness parameter returned by the thickness detector at the feeding point, which is 3.2 mm. Using the production speed of 192 and cardboard thickness of 3.2 mm as the combined operating condition identifier, a filtering operation is performed in the historical database. The historical database stores all operating records of this equipment over the past two years, each record containing a corresponding operating condition identifier field. After filtering, five candidate historical operating records with identical operating condition identifiers are obtained.

[0056] In S4, the second state feature vector is extracted from the current state feature sequence, with a maximum value of 16.1 microvolts, a minimum value of 12.3 microvolts, and an average value of 14.2 microvolts. Simultaneously, the second historical state feature vector is extracted from the first candidate historical record, with a maximum value of 15.8 microvolts, a minimum value of 11.9 microvolts, and an average value of 13.8 microvolts. The maximum value component is subtracted from the minimum value component, and the absolute value is taken to obtain 0.3. The minimum value component is subtracted from the minimum value component, and the absolute value is taken to obtain 0.4. The average value component is subtracted from the minimum value component, and the absolute value is taken to obtain 0.4. Then, 0.3 is added to the result. Adding 0.4 and 0.4 together gives a difference of 1.1 at that position. The same method is applied to all state feature vectors from the first to the 20th position in the sequence, accumulating the absolute values ​​of the differences to obtain a total difference of 23.7 between the current sequence and the first historical record. This process is repeated to calculate the total difference between the current sequence and the remaining four candidate historical records, yielding 18.9, 25.1, 20.4, and 33.2 respectively. The historical record corresponding to the smallest value of 18.9 is selected as the matching record.

[0057] In S5, the fault occurrence time recorded in the matching record is read, which is 8:30 AM on June 25, 2026. Simultaneously, the current host computer system time is read, which is 10:15 AM on July 4, 2026. The time interval between the two is calculated: nine days from June 25 to July 4, plus 2 hours and 45 minutes, resulting in 218 hours and 45 minutes. This value is used as the predicted remaining normal operating time. The equipment management terminal has a pre-set maintenance threshold of 192 hours. Since the predicted remaining time of 218 hours and 45 minutes is greater than 192 hours, it is determined that the current level has not reached the point where maintenance is necessary, and no maintenance prompt instruction is generated. If the subsequently calculated predicted remaining time is lower than this threshold, a text-formatted maintenance prompt instruction containing the equipment number, the predicted remaining time, and the matching record number is generated and output to the display interface of the workshop equipment management terminal via the industrial Ethernet.

[0058] Furthermore, the time window length in S2 is dynamically determined based on the periodic pulses in the vibration signal collected by the sensor node, including:

[0059] The pulse moments in the vibration time series where the amplitude of two adjacent peak values ​​exceeds the amplitude threshold are detected, the time interval between adjacent pulse moments is calculated as the equipment rotation period, and the length of the time window is set as an integer multiple of the rotation period.

[0060] In this embodiment, the determination of the time window length for the vibration time series involves the following dynamic calculation: In the original vibration signal collected by the sensor, an impact pulse is generated each rotation of the main shaft of the wire-feeding machine due to the bearing rolling element passing through the defect. This pulse exhibits a significant spike in vibration amplitude. The host computer program searches backward from the beginning of the vibration time series, setting an amplitude threshold of 14 microvolts. When the first amplitude point exceeding 14 microvolts is found, it is recorded as the first pulse moment, which is the 120th sampling point in the series, or 58.6 milliseconds. Continuing the search, the second amplitude point exceeding 14 microvolts is found and recorded as the second pulse moment, which is the 328th sampling point in the series, or 160.2 milliseconds. The difference between the second pulse moment and the first pulse moment is calculated to be 101.6 milliseconds, which is the current rotation cycle of the device spindle. To ensure that each time window contains at least one complete rotation cycle of vibration information, the time window length is set to four times the rotation cycle value, i.e., 406.4 milliseconds. Then, using this dynamically calculated 406.4 milliseconds as the actual time window length used in S2, the 10-second vibration time series is re-divided. The first time window starts from 0 milliseconds and ends at 406.4 milliseconds, the second time window starts from 406.4 milliseconds and ends at 812.8 milliseconds, and so on until the entire 10-second acquisition data is covered.

[0061] Furthermore, the difference accumulation calculation in S4 specifically involves:

[0062] Calculate the absolute values ​​of the differences between the current state feature vector and the corresponding state feature vector in the historical state feature sequence, including the absolute values ​​of the maximum, minimum, and average component differences. Then, add the three absolute values ​​of the differences at the same position to obtain the position difference value at that position. Finally, sum the position difference values ​​of all positions to obtain the total difference value.

[0063] In this embodiment, the calculation of the specific difference between the current state feature sequence and a certain historical state feature sequence at the same position is refined. Assume that the three components of the fifth state feature vector in the current state feature sequence are a maximum value of 17.2 microvolts, a minimum value of 13.1 microvolts, and an average value of 15.4 microvolts, while the three components of the fifth historical state feature vector in the historical state feature sequence are a maximum value of 16.5 microvolts, a minimum value of 12.8 microvolts, and an average value of 14.9 microvolts. First, the absolute value of the difference between the maximum value component and the historical state feature vector is calculated to be 0.7. Then, the absolute value of the difference between the minimum value component and the historical state feature vector is calculated to be 0.3. Finally, the absolute value of the difference between the average value component and the historical state feature vector is calculated to be 0.5. After calculating these three absolute value differences, they are added together to obtain the position difference value of 1.5 for the fifth position. Following the same procedure, calculate the positional difference values ​​for the first, second, third, and twentieth positions in the sequence. Assume the difference value for the first position is 0.8, the second is 1.2, the third is 0.9, the fourth is 1.1, and so on. Finally, sum all twenty positional difference values ​​to obtain the total difference value between the current state feature sequence and the historical state feature sequence. This total difference value serves as a quantitative measure of the similarity between the two sequences and is used for subsequent comparisons of numerical values ​​across different historical records.

[0064] Furthermore, referring to Figure 2 In step S4, when the length of the current state feature sequence is inconsistent with the length of the historical state feature sequence, the following length alignment operation is performed:

[0065] If the length of the current state feature sequence is less than the length of the historical state feature sequence, then a prefix subsequence with the same length as the current state feature sequence is extracted from the beginning of the historical state feature sequence, and this prefix subsequence is used as the historical state feature sequence to participate in the difference accumulation calculation.

[0066] If the length of the current state feature sequence is greater than the length of the historical state feature sequence, then the starting end of the current state feature sequence is used as the reference, and a prefix subsequence with the same length as the historical state feature sequence is extracted from the current state feature sequence as the current state feature sequence to participate in the difference accumulation calculation.

[0067] After length alignment is completed, the maximum value component, minimum value component, and average value component of each state feature vector in the aligned current state feature sequence are taken in chronological order. The absolute value of the difference is calculated with the maximum value component, minimum value component, and average value component of the corresponding position state feature vector in the aligned historical state feature sequence. The absolute values ​​of the differences of the three components at the same position are added together to obtain the difference component at that position. The difference components at all positions are then summed to obtain the total difference value.

[0068] For each historical running record in the historical database, the above length alignment and total difference value calculation are performed to obtain the total difference value between the current state feature sequence and each historical state feature sequence. Then, all total difference values ​​are sorted, and the historical running record corresponding to the historical state feature sequence with the smallest total difference value after sorting is determined as the matching record.

[0069] In this embodiment, when the length of the current state feature sequence is inconsistent with the length of a candidate historical state feature sequence in the historical database, the following detailed alignment and calculation operations are performed. Assume that the current state feature sequence, after being divided by S2, contains 30 state feature vectors, while the first historical running record in the historical database stores only 25 state feature vectors. Since the current sequence length of 30 is greater than the historical sequence length of 25, the host computer uses the beginning of the current state feature sequence as a reference, that is, starting from the first state feature vector in the sequence, sequentially truncating the first 25 state feature vectors, discarding the 26th to 30th state feature vectors. These truncated first 25 vectors constitute a temporary current sequence, used for subsequent difference accumulation. Suppose the second historical record in the historical database stores a historical state feature sequence containing 35 state feature vectors, while the current sequence length is 30. Since the current sequence length is shorter than the historical sequence length, starting from the beginning of this second historical state feature sequence (i.e., from the first historical state feature vector), the first 30 historical state feature vectors are extracted, discarding the 31st to 35th vectors. These 30 extracted vectors constitute a temporary historical sequence. After this length alignment, for the first historical record, the 25 vectors of the temporary current sequence are matched one-to-one with the 25 vectors of the first historical record for calculation. Taking the tenth position as an example, the maximum value of the tenth vector in the temporary current sequence is 18.2, the minimum value is 14.0, and the average value is 16.1. The maximum value of the tenth vector in the first historical record is 17.5, the minimum value is 13.3, and the average value is 15.4. The absolute values ​​of the differences are calculated to be 0.7, 0.7, and 0.7 respectively, resulting in a difference value of 2.1 for this position. The difference values ​​of the 25 positions are summed to obtain a total difference value of 48.3. For the second historical record, the 30 vectors of the current sequence and the 30 vectors of the temporary historical sequence are used to calculate the difference value one-to-one with the position. Taking the 15th position as an example, the difference value is calculated and then summed for 30 positions to obtain a total difference value of 52.7. There is also a third record in the historical database, whose historical state feature sequence length is 30, which is the same as the current sequence length. No truncation is needed, and the total difference value of 44.5 is calculated directly for 20 positions. After calculating the total difference value for all candidate records, a set of values ​​is obtained: 48.3, 52.7, 44.5, 62.1, and 39.8. The host computer sorts the values ​​in ascending order, and the sorting results are 39.8, 44.5, 48.3, 52.7, and 62.1. The historical record corresponding to the smallest value of 39.8 at the beginning of the sorted list, i.e., the third historical record, is selected as the final matching record.

[0070] Furthermore, the preset threshold in S5 is the average running time calculated based on the fault occurrence time node and equipment start running time node corresponding to all historical operation records with the same identifier as the current operating condition in the historical database. The average running time is multiplied by a preset proportional coefficient to obtain the value as the preset threshold.

[0071] In this embodiment, through the filtering of operating condition identifiers in S3, there are five candidate historical operation records in the historical database that have the same production speed of 192 and cardboard thickness of 3.2. The host computer reads the fault occurrence time node in the first candidate record as 10:00 AM on May 20, 2026, and simultaneously reads the equipment start-up time node in the same record as 8:00 AM on May 15, 2026. The calculation shows that the running time of this record from the start of operation to the fault occurrence is 5 days and 2 hours, or 122 hours. In the second candidate record, the fault occurrence time node is 2:30 PM on June 1, 2026, and the equipment start-up time node is 7:00 AM on May 26, 2026, with a running time of 6 days, 7 hours and 30 minutes, or 151.5 hours. In the third candidate record, the fault occurred at 9:15 AM on June 10, 2026, and the equipment started operating at 6:00 AM on June 3, 2026, with a running time of 7 days, 3 hours, and 15 minutes, or 171.25 hours. In the fourth candidate record, the fault occurred at 5:40 PM on June 18, 2026, and the equipment started operating at 8:30 AM on June 11, 2026, with a running time of 7 days, 9 hours, and 10 minutes, or 177.17 hours. In the fifth candidate record, the fault occurred at 8:30 AM on June 25, 2026, and the equipment started operating at 9:00 AM on June 18, 2026, with a running time of 6 days, 23 hours, and 30 minutes, or 167.5 hours. Adding the five runtimes together gives a total runtime of 789.42 hours. Dividing this by 5 gives an average runtime of 157.88 hours. This average runtime is the average cycle from the start of operation to the occurrence of a fault under the current operating conditions. The preset proportional coefficient is 0.8. Multiplying the average runtime of 157.88 hours by 0.8 gives a product of 126.30 hours. Rounding this product value is used as the preset threshold, i.e., the maintenance judgment threshold adopted by the equipment management terminal is 126 hours. If the predicted remaining normal operating time calculated in S5 is less than 126 hours, a maintenance prompt instruction will be triggered.

[0072] Furthermore, each historical operation record in the historical database in S3 also includes the total cumulative operating time of the equipment corresponding to the record. After filtering out historical operation records with the same identifier as the current operating condition from the historical database, the total cumulative operating time of the current equipment is obtained, and the absolute value of the difference between the total cumulative operating time of the current equipment and the total cumulative operating time of the equipment in each filtered historical operation record is calculated. A preset number of historical operation records with the smallest absolute value of the difference are selected from the filtered historical operation records, and the preset number of historical operation records are used as the objects for difference accumulation calculation in S4.

[0073] In this embodiment, after completing the screening of operating condition identifiers, a secondary screening operation based on the total cumulative operating time of the equipment is further performed. The host computer first reads the total cumulative operating time recorded by the internal timer of the current contact machine, which is 1872 hours. Then, from the five candidate historical operating records with the same operating condition identifier as the current one selected from the historical database, the total cumulative operating time of the equipment stored in each record is read. The total cumulative operating time of the first candidate record is 1920 hours, which is 48 hours less than the current cumulative operating time of 1872 hours. The total cumulative operating time of the second candidate record is 1805 hours, which is 67 hours less than the current operating time. The total cumulative operating time of the third candidate record is 1690 hours, which is 67 hours less than the current operating time. The difference between the lengths is 182 hours; the cumulative running time of the fourth candidate record is 1993 hours, and the difference between this and the current time is 121 hours; the cumulative running time of the fifth candidate record is 1854 hours, and the difference between this and the current time is 18 hours; the preset number is set to three, so the three records with the smallest absolute value are selected from these five differences, namely the fifth record with a difference of 18 hours, the first record with a difference of 48 hours, and the second record with a difference of 67 hours; after this second filtering, the original five candidate records are reduced to three, namely the fifth, the first, and the second; the subsequent difference accumulation calculation and minimum difference selection operation in S4 are only performed within these three records, and the third and fourth records with larger differences are no longer considered.

[0074] Furthermore, referring to Figure 3 The specific process of generating maintenance prompt instructions in S5 is as follows:

[0075] The fault type in the matching record is obtained as the expected fault type. At the same time, the fault occurrence time node in the matching record is obtained. The difference between the fault occurrence time node and the current time node is calculated to obtain the predicted remaining normal operation time.

[0076] The predicted remaining normal operating time is compared with multiple pre-divided time intervals to determine the time interval to which the predicted remaining normal operating time belongs. Each time interval corresponds to a pre-set maintenance urgency level. Based on the determined maintenance urgency level, the corresponding maintenance operation description text is extracted from the maintenance measures table. The maintenance measures table stores the correspondence between different maintenance urgency levels and maintenance operation description texts.

[0077] The expected fault type, the maintenance urgency level, and the maintenance operation description text are combined into structured maintenance prompt data;

[0078] After generating maintenance prompts and outputting them to the equipment management terminal, the vibration time sequence at the current moment, the current state characteristic sequence, the identifier of the matching record, and the current time node are also stored as new historical operation records in the historical database.

[0079] The new historical operation record marks the status of the device as not yet having failed. When the device actually fails later, the actual failure time and the actual failure type will be updated to the historical operation record.

[0080] In this embodiment, assuming the fault type recorded in the matching record is a broken drive belt, the host computer reads the fault occurrence time node in the matching record as 8:30 AM on June 25, 2026, and the current time node is 10:15 AM on July 4, 2026. The difference is calculated to obtain a predicted remaining normal operation time of 218 hours and 45 minutes. The host computer internally divides the time into four time intervals: the first interval is 0 to 12 hours, the second interval is 12 to 48 hours, the third interval is 48 to 120 hours, and the fourth interval is more than 120 hours. The current predicted remaining time of 218 hours and 45 minutes belongs to the fourth interval, that is, more than 120 hours. Each time interval corresponds to a maintenance emergency level: the first interval corresponds to the first level of emergency, the second interval corresponds to the second level of emergency, the third interval corresponds to the third level of emergency, and the fourth interval corresponds to the fourth level of normal emergency. Therefore, the current maintenance emergency level is determined to be level four. The host computer internally stores a maintenance measures table, indexed by maintenance urgency level. Level four corresponds to the maintenance operation description text: "Inspect the drive belt surface for cracks and record the belt tension value." The host computer combines three information fields—the fault type (drive belt wear and breakage), the determined maintenance urgency level (Level 4), and the operation description text extracted from the maintenance measures table ("Inspect the drive belt surface for cracks and record the belt tension value")—into a structured maintenance prompt data packet. This data packet is generated in text form and includes the equipment name (wire contact machine), equipment number (M-07), predicted remaining time (218 hours and 45 minutes), expected fault type, urgency level, and operation description. This maintenance prompt data packet is then output to the device management terminal's display screen via industrial Ethernet. After the output action is completed, the host computer appends the entire vibration time series collected at the current moment, the current state feature sequence composed of all twenty state feature vectors calculated by S2, the unique ID R-2026-025 of this matching record in the historical database, and the current time node July 4, 2026, at 10:15 AM as a new historical operation record to the historical database. When storing data, a null value is entered in the fault occurrence time field and the fault type field of the new record, and a "no fault occurred" flag is specifically written in the status flag field. When the wire-connecting machine actually experiences a transmission belt breakage fault during subsequent operation, the maintenance personnel enter the actual fault occurrence time (8:30 PM on July 4, 2026) and the actual fault type (transmission belt breakage) into the equipment management terminal. The host computer retrieves the previously stored record based on the equipment number and status flag, updates the null value in the time field to the actual fault occurrence time, updates the null value in the fault type field to the actual fault type, and changes the status flag field from "never fault occurred" to "fault occurred".

[0081] Furthermore, in the difference accumulation calculation process in S4, different weighting coefficients are assigned to the maximum value component, the minimum value component, and the average value component, specifically as follows:

[0082] Obtain the production speed parameters in the current S3, and determine the weight configuration based on the speed range in which the production speed parameters are located. The weight configuration specifies the weight values ​​of the maximum value component, the minimum value component, and the average value component.

[0083] When calculating the difference components at each location, the absolute value of the difference of the maximum value component is multiplied by the weight of the maximum value component to obtain the weighted maximum value difference; the absolute value of the difference of the minimum value component is multiplied by the weight of the minimum value component to obtain the weighted minimum value difference; the absolute value of the difference of the average value component is multiplied by the weight of the average value component to obtain the weighted average value difference; then the weighted maximum value difference, weighted minimum value difference, and weighted average value difference are added together to obtain the difference component at that location; finally, the difference components at all locations are summed to obtain the difference value.

[0084] In this embodiment, the current production speed parameter obtained by the host computer in S3 is 192 cartons per minute. The host computer internally divides the production speed into three intervals: a low-speed interval (0 to 100 cartons per minute), a medium-speed interval (101 to 160 cartons per minute), and a high-speed interval (161 to 250 cartons per minute); the current speed of 192 falls into the high-speed interval. Based on the correspondence table between speed intervals and weight configurations, the host computer determines the weight configuration for the high-speed interval as follows: maximum value component weight value 0.6, minimum value component weight value 0.2, and average value component weight value 0.2. When calculating the difference component between the current state feature sequence and a historical state feature sequence at the first position, the following steps are taken: First, obtain the maximum value (18.5), minimum value (14.2), and average value (16.8) of the first vector of the current sequence, and the maximum value (17.3), minimum value (13.5), and average value (15.9) of the first vector of the historical sequence. Calculate the absolute value difference of 1.2, multiply it by the maximum value component weight value of 0.6, and obtain a weighted maximum value difference of 0.72. Calculate the absolute value difference of 0.7, multiply it by the minimum value component weight value of 0.2, and obtain a weighted minimum value difference of 0.14. Calculate the absolute value difference of 0.9, multiply it by the average value component weight value of 0.2, and obtain a weighted average value difference of 0.18. Then, add the weighted maximum value difference of 0.72, the weighted minimum value difference of 0.14, and the weighted average value difference of 0.18 to obtain the difference component at the first position, which is 1.04. Subsequent positions from the 2nd to the 20th position are calculated using the same weighted values, and the difference components at each position are then summed to obtain the total difference value. If the current production speed falls into the medium speed range, the weight configuration will switch to the maximum value weight of 0.4, the minimum value weight of 0.3, and the average value weight of 0.3; if it falls into the low speed range, the weight configuration will switch to the maximum value weight of 0.2, the minimum value weight of 0.4, and the average value weight of 0.4. The sum of the weight values ​​in different speed ranges will always be equal to 1.0.

[0085] Furthermore, in step S5, after generating the maintenance prompt command and outputting it to the device management terminal, maintenance confirmation is also performed:

[0086] During the preset waiting period after the output maintenance prompt command, vibration signals continue to be collected through the sensor node, and the difference between the current state feature sequence and each historical state feature sequence in the historical database is recalculated according to the methods of S2 to S4. Matching records are reselected, and the remaining normal operation time is recalculated based on the reselected matching records.

[0087] If the recalculated predicted remaining normal operating time is still lower than the preset threshold, the previously output maintenance prompt instruction remains unchanged; if the recalculated predicted remaining normal operating time is not lower than the preset threshold, a cancellation instruction is sent to the device management terminal to cancel the previously output maintenance prompt instruction.

[0088] In this embodiment, it is assumed that the host computer outputs a maintenance prompt command at 10:15 AM on July 4, 2026, predicting that the remaining normal operating time is 85 hours, which is lower than the preset threshold of 126 hours. After outputting the command, the host computer starts a timer with a preset waiting period of 30 minutes. When the timer reaches 30 minutes, i.e., 10:45 AM, the host computer re-collects a new 10-second continuous vibration signal from 10:45 AM via the sensor, and divides the new vibration signal into 20 new time windows according to method S2, extracting the maximum, minimum, and average values ​​of each window to form a new current state feature sequence; then, according to method S3, it re-acquires the current production speed parameters and raw material thickness parameters. If the parameters have not changed, the original working condition identifier is used to filter the historical database; according to method S4, the difference between the new current state feature sequence and each filtered historical state feature sequence in the historical database is recalculated, and a new matching record is selected; assuming that the newly selected matching record is the same as before, its fault occurrence time node is still June 25, 2026. At 8:30 AM, the current recalculation time is 10:45 AM, and the recalculated remaining normal operation time is 83 hours and 15 minutes. This recalculated 83 hours and 15 minutes is still lower than the preset threshold of 126 hours. Therefore, the host computer maintains the previously output maintenance prompt instruction unchanged and does not perform any cancellation operation. The device management terminal continues to display the maintenance prompt. In another scenario, if the recalculated predicted remaining normal operation time becomes 130 hours, which is not lower than the preset threshold of 126 hours, the host computer generates a cancellation instruction. This cancellation instruction contains the generation timestamp and instruction number of the original maintenance prompt instruction and is sent to the device management terminal via Ethernet. After receiving the cancellation instruction, the device management terminal removes the previously displayed maintenance prompt information from the screen and records the cancellation log.

[0089] Furthermore, in S3, the historical operation records in the historical database are divided into multiple fault category sub-databases according to fault type, and each fault category sub-database stores a sequence of historical state features with the same fault type;

[0090] Before selecting the historical state feature sequence with the smallest difference in S4, the trend of change of the maximum value component and the average value component in the current state feature sequence is determined. The trend of change is determined as follows:

[0091] Take the state feature vectors of the last three time windows in the current state feature sequence, and calculate the successive difference signs of the maximum value component and the average value component in these three time windows respectively. If the successive difference signs of the maximum value component and the average value component are both positive, the trend direction is determined to be an upward trend; if both are negative, it is determined to be a downward trend; otherwise, it is determined to be a stable trend. After obtaining the trend direction of the current state feature sequence, candidate records with the same trend direction as its historical state feature sequence are selected from the historical database. The trend direction of the historical state feature sequence is pre-stored in the historical database. Then, the difference accumulation calculation and minimum difference selection of S4 are performed on the candidate records.

[0092] In this embodiment, a fault category sub-database division and trend direction judgment operation are introduced. The historical database is physically divided into four independent sub-databases according to different fault types. The first sub-database stores historical state feature sequences for bearing pitting fault type; the second sub-database stores historical state feature sequences for gear tooth breakage fault type; the third sub-database stores historical state feature sequences for drive belt aging fault type; and the fourth sub-database stores historical state feature sequences for rubber roller wear fault type. Before executing S4 to select the historical state feature sequence with the smallest difference value, the host computer judges the trend of change of the current state feature sequence. The last three time windows in the current state feature sequence, namely the eighteenth, nineteenth, and twentieth windows, are selected, each corresponding to a state feature vector. The maximum value component of the eighteenth window is 15.2 microvolts, and the average value component is 13.8 microvolts; the maximum value component of the nineteenth window is 16.8 microvolts, and the average value component is 14.5 microvolts; and the maximum value component of the twentieth window is 18.5 microvolts, and the average value component is 15.9 microvolts. For the maximum value component, the difference between the maximum value of the nineteenth window (16.8) and the maximum value of the eighteenth window (15.2) is calculated to obtain a difference of +1.6, and the difference between the maximum value of the twentieth window (18.5) and the maximum value of the nineteenth window (16.8) is calculated to obtain a difference of +1.7. Both differences are positive. For the average value component, the difference between the average value of the nineteenth window (14.5) and the average value of the eighteenth window (13.8) is calculated to obtain a difference of +0.7, and the difference between the average value of the twentieth window (15.9) and the average value of the twentieth window (14.5) is calculated to obtain a difference of +1.4. Both differences are also positive. Since the successive differences of the maximum value component and the average value component are all positive, the host computer determines that the trend of the current state feature sequence is upward. After the determination, the host computer only selects candidates from the historical operation records in the historical database that have a pre-stored trend direction field and that field is marked as upward. Each historical state feature sequence pre-stored in the historical database has its own trend direction calculated and recorded using the same method when it is entered into the database. After the selection, a total of twelve candidate records are obtained, which are distributed in different fault category sub-databases. The host computer ignores other records in the historical database that are marked as a downward trend or a stable trend, and only performs the difference accumulation calculation and minimum difference selection operation described in S4 on the above twelve candidate records. Finally, it selects the record with the smallest difference value from these candidate records as the matching record.

[0093] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0094] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0095] The specific embodiments of the invention have been described in detail above, but they are only examples, and this application is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions to the invention are also within the scope of this application. Therefore, all equivalent changes, modifications, and improvements made without departing from the spirit and principles of this application should be covered within the scope of this application.

Claims

1. A method for predictive maintenance of cardboard box production equipment based on sensor networks, characterized in that, include: S1, by continuously collecting vibration signals during the operation of the equipment through sensor nodes deployed in key parts of the carton and paper box production equipment, the vibration amplitude values ​​acquired at each sampling moment are arranged in the order of collection time to generate a vibration time series; S2, the vibration time series is divided into multiple consecutive time windows according to a preset time window length, the maximum value, minimum value and average value of vibration amplitude are extracted in each time window, and the maximum value, minimum value and average value in the same time window are combined into the state feature vector corresponding to the time window to form a state feature sequence; S3, obtain the current production speed parameters and raw material thickness parameters of the cardboard box production equipment, use the production speed parameters and raw material thickness parameters as working condition identifiers, filter out historical operating records that are the same as the current working condition identifiers from the historical database, each historical operating record contains a historical state feature sequence and the corresponding fault occurrence time node and fault type; S4, calculate the difference between the corresponding elements of the current state feature sequence and the historical state feature sequence in each historical running record by summing the differences between the current state feature sequence and each historical state feature sequence, and select the historical running record corresponding to the historical state feature sequence with the smallest difference as the matching record; S5. Based on the fault occurrence time node and the current time node in the matching record, calculate the time interval between the fault occurrence time node and the current time node, and use this time interval as the predicted remaining normal operation time. When the predicted remaining normal operation time is lower than the preset threshold, generate a maintenance prompt instruction and output it to the device management terminal.

2. The method according to claim 1, characterized in that, The time window length in S2 is dynamically determined based on the periodic pulses in the vibration signal collected by the sensor node, including: The pulse moments in the vibration time series where the amplitude of two adjacent peak values ​​exceeds the amplitude threshold are detected, the time interval between adjacent pulse moments is calculated as the equipment rotation period, and the length of the time window is set as an integer multiple of the rotation period.

3. The method according to claim 1, characterized in that, The difference accumulation calculation in S4 is specifically as follows: Calculate the absolute values ​​of the differences between the current state feature vector and the corresponding state feature vector in the historical state feature sequence, including the absolute values ​​of the maximum, minimum, and average component differences. Then, add the three absolute values ​​of the differences at the same position to obtain the position difference value at that position. Finally, sum the position difference values ​​of all positions to obtain the total difference value.

4. The method according to claim 1, characterized in that, In step S4, when the length of the current state feature sequence is inconsistent with the length of the historical state feature sequence, the following length alignment operation is performed: If the length of the current state feature sequence is less than the length of the historical state feature sequence, then a prefix subsequence with the same length as the current state feature sequence is extracted from the beginning of the historical state feature sequence, and this prefix subsequence is used as the historical state feature sequence to participate in the difference accumulation calculation. If the length of the current state feature sequence is greater than the length of the historical state feature sequence, then the starting end of the current state feature sequence is used as the reference, and a prefix subsequence with the same length as the historical state feature sequence is extracted from the current state feature sequence as the current state feature sequence to participate in the difference accumulation calculation. After length alignment is completed, the maximum value component, minimum value component, and average value component of each state feature vector in the aligned current state feature sequence are taken in chronological order. The absolute value of the difference is calculated with the maximum value component, minimum value component, and average value component of the corresponding position state feature vector in the aligned historical state feature sequence. The absolute values ​​of the differences of the three components at the same position are added together to obtain the difference component at that position. The difference components at all positions are then summed to obtain the total difference value. For each historical running record in the historical database, the above length alignment and total difference value calculation are performed to obtain the total difference value between the current state feature sequence and each historical state feature sequence. Then, all total difference values ​​are sorted, and the historical running record corresponding to the historical state feature sequence with the smallest total difference value after sorting is determined as the matching record.

5. The method according to claim 1, characterized in that, The preset threshold in S5 is the average running time calculated based on the fault occurrence time node and equipment start running time node corresponding to all historical operation records with the same identifier as the current operating condition in the historical database. The average running time is multiplied by a preset proportional coefficient to obtain the value as the preset threshold.

6. The method according to claim 1, characterized in that, Each historical operation record in the historical database in S3 also includes the total cumulative operating time of the equipment corresponding to the record. After filtering out historical operation records with the same identifier as the current operating condition from the historical database, the total cumulative operating time of the current equipment is obtained. The absolute value of the difference between the total cumulative operating time of the current equipment and the total cumulative operating time of the equipment in each filtered historical operation record is calculated. A preset number of historical operation records with the smallest absolute value of the difference are selected from the filtered historical operation records. Then, the preset number of historical operation records are used as the objects to participate in the difference accumulation calculation in S4.

7. The method according to claim 1, characterized in that, The specific process for generating maintenance prompt instructions in S5 is as follows: The fault type in the matching record is obtained as the expected fault type. At the same time, the fault occurrence time node in the matching record is obtained. The difference between the fault occurrence time node and the current time node is calculated to obtain the predicted remaining normal operation time. The predicted remaining normal operating time is compared with multiple pre-divided time intervals to determine the time interval to which the predicted remaining normal operating time belongs. Each time interval corresponds to a pre-set maintenance urgency level. Based on the determined maintenance urgency level, the corresponding maintenance operation description text is extracted from the maintenance measures table. The maintenance measures table stores the correspondence between different maintenance urgency levels and maintenance operation description texts. The expected fault type, the maintenance urgency level, and the maintenance operation description text are combined into structured maintenance prompt data; After generating maintenance prompts and outputting them to the equipment management terminal, the vibration time sequence at the current moment, the current state characteristic sequence, the identifier of the matching record, and the current time node are also stored as new historical operation records in the historical database. The new historical operation record marks the status of the device as not yet having failed. When the device actually fails later, the actual failure time and the actual failure type will be updated to the historical operation record.

8. The method according to claim 1, characterized in that, In the difference accumulation calculation process described in S4, different weighting coefficients are assigned to the maximum value component, the minimum value component, and the average value component, specifically as follows: Obtain the production speed parameters in the current S3, and determine the weight configuration based on the speed range in which the production speed parameters are located. The weight configuration specifies the weight values ​​of the maximum value component, the minimum value component, and the average value component. When calculating the difference components at each location, the absolute value of the difference of the maximum value component is multiplied by the weight of the maximum value component to obtain the weighted maximum value difference; the absolute value of the difference of the minimum value component is multiplied by the weight of the minimum value component to obtain the weighted minimum value difference; the absolute value of the difference of the average value component is multiplied by the weight of the average value component to obtain the weighted average value difference; then the weighted maximum value difference, weighted minimum value difference, and weighted average value difference are added together to obtain the difference component at that location; finally, the difference components at all locations are summed to obtain the difference value.

9. The method according to claim 1, characterized in that, In step S5, after generating the maintenance prompt command and outputting it to the device management terminal, maintenance confirmation is also performed: During the preset waiting period after the output maintenance prompt command, vibration signals continue to be collected through the sensor node, and the difference between the current state feature sequence and each historical state feature sequence in the historical database is recalculated according to the methods of S2 to S4. Matching records are reselected, and the remaining normal operation time is recalculated based on the reselected matching records. If the recalculated predicted remaining uptime is still lower than the preset threshold, the output maintenance prompt instruction will remain unchanged. If the recalculated predicted remaining normal operating time is not lower than the preset threshold, a cancellation command is sent to the device management terminal to cancel the previously output maintenance prompt command.

10. The method according to claim 1, characterized in that, In S3, the historical operation records in the historical database are divided into multiple fault category sub-databases according to the fault type. Each fault category sub-database stores a sequence of historical state features with the same fault type. Before selecting the historical state feature sequence with the smallest difference in S4, the trend of change of the maximum value component and the average value component in the current state feature sequence is determined. The trend of change is determined as follows: Take the state feature vectors of the last three time windows in the current state feature sequence, and calculate the successive difference sign of the maximum value component and the successive difference sign of the average value component in these three time windows respectively. If the successive difference signs of the maximum value component and the average value component are both positive, the trend direction is determined to be an upward trend. If they are both negative, it is determined to be a downward trend. Otherwise, it is determined to be a stable trend. After obtaining the trend direction of the current state feature sequence, candidate records with the same trend direction as its historical state feature sequence are selected from the historical database. The trend direction of the historical state feature sequence is stored in the historical database in advance. Then, the difference accumulation calculation and minimum difference selection of S4 are performed in the candidate records.

Images