A smart enterprise management platform based on big data

Abstract

This invention relates to the field of big data technology, specifically disclosing a smart enterprise management platform based on big data, comprising the following modules: a window sampling module: acquiring the root mean square of vibration and operating conditions through a preset sampling window, and classifying similar operating conditions; a suspicious candidate module: constructing a list of vibrations under the same operating conditions, calculating quantiles to obtain anomaly coefficients; a horizontal comparison module: filtering equipment of the same type and operating conditions, sorting and calculating horizontal anomaly coefficients; and a fault prediction module: calculating the ideal period based on the average difference of historical fault intervals, and sending fault warnings at regular intervals. This invention addresses the difficulty in identifying early degradation of compressors under strong noise and variable operating conditions. Through operating condition stratification, vertical self-comparison, and horizontal group comparison, it captures subtle anomalies and achieves pre-fault warnings based on historical fault intervals, effectively reducing the risk of unplanned downtime, promoting the shift from periodic maintenance to predictive maintenance, and avoiding under- or over-maintenance.

Description

Technical Field

[0001] This invention relates to the field of big data technology, and more specifically to a smart enterprise management platform based on big data. Background Technology

[0002] In modern manufacturing, unplanned compressor downtime is the primary risk affecting the stability of gas supply to production lines and overall delivery capacity. To mitigate sudden failures, companies generally adopt reactive maintenance or periodic maintenance strategies. Reactive maintenance directly leads to production interruptions and may cause secondary losses such as liquid carryover in compressed gas and unit damage; periodic maintenance, on the other hand, suffers from the dual drawbacks of insufficient or excessive maintenance.

[0003] With the popularization of Industrial Internet of Things (IIoT) technology, enterprises have begun to deploy vibration, temperature, and other condition monitoring sensors in key parts of compressors, attempting to achieve predictive maintenance through data-driven methods. However, in practical engineering applications, a key technical bottleneck has remained unresolved: how to accurately identify subtle signs of early compressor degradation from high-concurrency time-series data in environments with high noise and variable operating conditions, and provide reliable early warnings at the nascent stage of a fault. During normal compressor operation, vibration and temperature signals are often masked by noise from airflow pulsations, valve impacts, environmental vibrations, and fluctuations in operating conditions. Early abnormal characteristics are extremely weak, and traditional fixed threshold methods are prone to missed or false alarms. These problems mean that predictive maintenance for compressors in most enterprises remains at the stage of post-event alarms or simple threshold monitoring, failing to truly achieve the goal of intelligent management with early warnings before faults occur. Summary of the Invention

[0004] The purpose of this invention is to provide a smart enterprise management platform based on big data to solve the above-mentioned technical problems.

[0005] The objective of this invention can be achieved through the following technical solutions: A smart enterprise management platform based on big data includes the following modules: Window sampling module: Preset sampling windows, and after each sampling window is completed, obtain the root mean square vibration of the compressor, the average rotor speed V, and the average load F within that window; construct the operating condition combination of the compressor in that sampling window based on the average rotor speed V and the average load F, and record all sampling windows with the same operating condition combination as the same operating condition. Suspicious candidate modules: For each compressor, obtain the root mean square of vibration under the same operating conditions in sequence, and construct a list of root mean square vibration under the same operating conditions [x1, x2, ..., xi], where xi represents the root mean square of vibration in the i-th sampling window under the same operating conditions; Obtain the root mean square (RMS) X of the current sampling window and the corresponding list of RMS vibrations, and then obtain the λ quantile from the RMS vibration list. and γ quantile , where λ represents the preset first coefficient and γ represents the preset second coefficient; Based on λ quantiles and γ quantile Calculate the anomaly coefficient S of the compressor in the current sampling window; Horizontal comparison module: If S is greater than or equal to the preset screening threshold The compressor is designated as a suspected candidate. Within the same workshop, compressors of the same model and with the same current operating condition are selected. The root mean square (RMS) vibration of these compressors within the most recent complete sampling window is obtained. Based on the obtained RMS vibration data, the transverse anomaly coefficient corresponding to the suspected candidate is calculated. ; Fault prediction module: If the horizontal anomaly coefficient If the value is greater than or equal to the preset threshold for judging true anomalies, the corresponding suspicious candidate is recorded as a true anomaly. The mean T and standard deviation s of the historical fault intervals of the true anomaly are obtained, the ideal fault period Ts = T - 2s is calculated, and the time of the most recent fault of the true anomaly is obtained. When at time Send compressor fault warnings in real time.

[0006] As a further aspect of the present invention: in the aforementioned window sampling module, constructing the compressor's operating condition combination within the sampling window based on the average rotor speed V and average load F specifically includes the following steps: Calculate the RPM gear and load settings Where v represents the preset basic speed and f represents the preset basic load, the operating condition combination of the compressor in this sampling window is generated. .

[0007] As a further aspect of the present invention: in the aforementioned suspected candidate module, the first coefficient λ satisfies 90%≤λ≤95%, and the second coefficient γ satisfies γ≥99.9%.

[0008] As a further aspect of the present invention: in the aforementioned suspected candidate module, based on the λ quantile... and γ quantile The method for calculating the anomaly coefficient S of the compressor in the current sampling window specifically includes the following steps: Calculate the anomaly coefficient of the compressor within the current sampling window. ,in, This represents the preset judgment threshold.

[0009] As a further aspect of the present invention: in the lateral comparison module, the lateral anomaly coefficient corresponding to the suspected candidate is calculated based on the obtained root mean square vibration. Specifically, the following steps are included: The obtained root mean square (RMS) vibrations are sorted in ascending order, and the ascending rank r of the RMS vibration X corresponding to the suspected candidate is obtained. The transverse anomaly coefficient is then calculated. N represents the total number of elements in the sorting process. This represents the median of the root mean square of the vibration.

[0010] As a further aspect of the present invention: in the horizontal comparison module, the number of compressors that are the same as the suspected candidate models and have the same current operating condition combination is counted. When the number is less than the preset judgment number, subsequent operations are stopped and no fault time prediction is performed.

[0011] As a further aspect of the present invention: in the fault prediction module, if the total working time of the compressor recorded as a real abnormality is less than or equal to the preset aging working time, and the number of faults is less than or equal to 3, then no compressor fault warning will be issued.

[0012] As a further aspect of the present invention: in the window sampling module, if the sampling window contains a compressor shutdown period, the data obtained from that sampling window is discarded.

[0013] The beneficial effects of this invention are as follows: The intelligent enterprise management platform constructed in this solution addresses the technical bottleneck of compressors exhibiting weak early-stage degradation characteristics under strong noise and variable operating conditions, which are easily masked by operating condition fluctuations and background interference. It proposes a full-chain anomaly detection and fault early warning mechanism, encompassing data cleaning, operating condition stratification, longitudinal self-comparison, and lateral group comparison. By dividing the root mean square (RMS) of vibration into levels based on speed and load, a list of RMS vibrations under the same operating conditions is constructed, effectively eliminating the influence of operating condition changes on vibration amplitude and making the analyzed data more comparable. The anomaly coefficient is calculated using high quantiles and very high quantiles, capturing subtle changes in the RMS vibration from the tail end of the normal distribution towards extreme values, avoiding false alarms or missed alarms caused by fixed thresholds during operating condition switching. When longitudinal anomalies exceed the screening threshold, a lateral comparison is further introduced within the same workshop, of the same model, and under the same operating conditions. The lateral anomaly coefficient is calculated by the ascending ranking of the RMS vibrations and the median, effectively eliminating interference from individual equipment differences and common environmental noise, thus improving the reliability of anomaly identification. For compressors confirmed to be truly abnormal, the ideal failure cycle is calculated based on historical failure interval statistics, and an early warning is issued by combining this with the time of the most recent failure, enabling proactive prediction of the next failure's timing. Overall, this platform can accurately identify early signs of degradation from high-concurrency, high-noise monitoring data, significantly reducing the risk of unplanned downtime. It guides enterprises to shift from scheduled maintenance to predictive maintenance based on equipment health status, avoiding both production interruptions caused by insufficient maintenance and resource waste from excessive maintenance, truly achieving the engineering goal of early warning of compressor failures. Attached Figure Description

[0014] The invention will now be further described with reference to the accompanying drawings.

[0015] Figure 1 This is a schematic diagram of the structure of a smart enterprise management platform based on big data according to the present invention; Figure 2 This is a flowchart illustrating a smart enterprise management platform based on big data according to the present invention. Detailed Implementation

[0016] 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.

[0017] Please see Figure 1 As shown, this invention is a smart enterprise management platform based on big data, comprising the following modules: Window sampling module: Preset sampling windows, and after each sampling window is completed, obtain the root mean square vibration of the compressor, the average rotor speed V, and the average load F within that window; construct the operating condition combination of the compressor in that sampling window based on the average rotor speed V and the average load F, and record all sampling windows with the same operating condition combination as the same operating condition. Suspicious candidate modules: For each compressor, obtain the root mean square of vibration under the same operating conditions in sequence, and construct a list of root mean square vibration under the same operating conditions [x1, x2, ..., xi], where xi represents the root mean square of vibration in the i-th sampling window under the same operating conditions; Obtain the root mean square (RMS) X of the current sampling window and the corresponding list of RMS vibrations, and obtain the λ quantiles in the RMS vibration list. and γ quantile , where λ represents the preset first coefficient and γ represents the preset second coefficient; Based on λ quantiles and γ quantile Calculate the anomaly coefficient S of the compressor in the current sampling window; Horizontal comparison module: If S is greater than or equal to the preset screening threshold The compressor is designated as a suspected candidate. Within the same workshop, compressors of the same model and with the same current operating condition are selected. The root mean square (RMS) vibration of these compressors within the most recent complete sampling window is obtained. Based on the obtained RMS vibration data, the transverse anomaly coefficient corresponding to the suspected candidate is calculated. ; Fault prediction module: If the horizontal anomaly coefficient If the value is greater than or equal to the preset threshold for judging true anomalies, the corresponding suspicious candidate is recorded as a true anomaly. The mean T and standard deviation s of the historical fault intervals of the true anomaly are obtained, the ideal fault period Ts = T - 2s is calculated, and the time of the most recent fault of the true anomaly is obtained. When at time Send compressor fault warnings in real time.

[0018] It should be noted that the compressor operating data is first normalized using a window sampling module. The system has a preset fixed sampling window duration, and after each window ends, it records the root mean square of compressor vibration, average rotor speed, and average load for that period. To avoid interference from shutdown periods, if a window includes a compressor shutdown process, the entire data for that window is discarded. For the remaining sampling windows, the average speed and average load are divided into levels based on preset base speed and base load. The speed level is calculated by dividing the average speed by the base speed and rounding down, and the load level is calculated by dividing the average load by the base load and rounding down. This generates operating condition combinations, and all windows with the same operating condition combination are grouped into the same operating condition. In this way, subsequent analysis is only performed within the same operating condition, effectively eliminating the interference of speed and load changes on vibration amplitude.

[0019] In the suspected candidate module, for each compressor, the root mean square (RMS) vibration data for all windows under the same operating condition is collected chronologically to form a historical list of RMS vibration data for that operating condition. Once the RMS vibration data for the latest sampling window is obtained, the λ quantile and γ quantile are calculated from the historical list for the same operating condition. These two quantiles are used to calculate the anomaly coefficient. When S is greater than or equal to a preset screening threshold, the compressor is marked as a suspected candidate.

[0020] Subsequently, the horizontal comparison module intervenes. Within the same workshop, all compressors with the same model and completely identical current operating conditions as the suspected candidate are screened, and their root mean square (RMS) vibration values ​​within the most recent complete sampling window are extracted. To ensure statistical validity, if the number of compressors meeting the criteria is less than the preset judgment number, subsequent operations are stopped, and no fault prediction is performed. When the number meets the requirement, these RMS vibration values ​​are sorted in ascending order, and the ascending rank of the suspected candidate's RMS vibration value and the median of the sequence are recorded. If the RMS vibration value of a suspected candidate is less than the median, the horizontal anomaly coefficient is directly set to 0; if it is greater than or equal to the median, the horizontal anomaly coefficient is calculated by subtracting 1 from the rank and dividing by the total number of elements by 1. The closer this coefficient is to 1, the more significantly the vibration level of the suspected candidate is higher than other compressors in the same workshop under the same operating conditions and model background.

[0021] Finally, the system proceeds to the fault prediction module. If the horizontal anomaly coefficient is greater than or equal to the preset true anomaly judgment threshold, the corresponding suspicious candidate is confirmed as a true anomaly. The system obtains the compressor's historical fault records, calculates the mean and standard deviation of adjacent fault intervals, and then obtains the ideal fault cycle as the mean minus twice the standard deviation. Simultaneously, the time of the most recent fault occurrence is recorded. When the system clock reaches the moment when the most recent fault time plus the ideal fault cycle, a compressor fault warning is sent. Furthermore, if the total operating time of the compressor recorded as a true anomaly is less than or equal to the preset aging operating time and the number of faults does not exceed 3, a warning may be temporarily withheld to avoid generating invalid alarms in the early, occasional fault stage. Through the above-mentioned combination of hierarchical comparison and statistical prediction, the platform can accurately capture early signs of compressor degradation under strong noise and variable operating conditions, achieving intelligent early warning before faults occur.

[0022] In another preferred embodiment of the present invention, constructing the compressor's operating condition combination within the sampling window based on the average rotor speed V and average load F specifically includes the following steps: Calculate the RPM gear and load settings Where v represents the preset basic speed and f represents the preset basic load, the operating condition combination of the compressor in this sampling window is generated. .

[0023] It is worth noting that the core significance of this solution lies in transforming the continuously changing speed and load of the compressor into discrete gear combinations, thereby generating reproducible operating condition combinations. Through gear division, the root mean square value of vibration, which was previously impossible to directly compare due to fluctuations in speed and load, can be compared longitudinally and laterally within the same operating condition combination. This effectively eliminates the interference of varying operating conditions on vibration amplitude, separating early signs of degradation from strong noise and operating condition fluctuations. This provides a stable benchmark for subsequent calculation of anomaly coefficients based on quantiles and lateral comparisons between equipment in the same workshop, improving the accuracy of fault early warning.

[0024] In another preferred embodiment of the present invention, the first coefficient λ satisfies 90%≤λ≤95%, and the second coefficient γ satisfies γ≥99.9%.

[0025] Understandably, the λ quantile represents the upper boundary of normal vibration of the compressor under the same operating conditions, while the γ quantile represents the threshold of extreme anomalies. The combination of these two can capture subtle changes in vibration drift from the tail of the normal distribution to extreme values, while avoiding false alarms caused by normal fluctuations or noise interference, thus accurately identifying early signs of degradation in noisy environments.

[0026] In another preferred embodiment of the invention, based on λ quantiles and γ quantile The method for calculating the anomaly coefficient S of the compressor in the current sampling window specifically includes the following steps: Calculating the anomaly coefficient S of the compressor in the current sampling window ,in, This represents the preset judgment threshold.

[0027] It is important to note that, through normalization and amplitude limiting, the current root mean square X of the vibration is compared to the normal upper boundary in the historical distribution. Boundaries of extreme anomalies The relative position, mapped to Within the interval. When When the molecule is non-positive, the result is 0, indicating that the current vibration is completely within the normal fluctuation range and no abnormal coefficient is generated; when When the ratio is greater than or equal to 1, it is then limited to... This indicates that the level of anomaly has been reached or exceeded; when X is between and When the anomaly coefficient S is between 0 and... The linear growth between these values ​​reflects the continuous process of vibration gradually deteriorating from the typical upper boundary to the extreme anomaly.

[0028] This formula has three significant advantages: First, it eliminates the differences in absolute vibration amplitude between different operating conditions and different equipment, making the degree of anomaly comparable; second, it mitigates the influence of extreme values ​​in nonlinear compression, avoiding coefficient saturation caused by single noise spikes; and third, it uses an upper limit... This ensures the stability of subsequent threshold screening. The formula can sensitively capture subtle signs of degradation in the vibration distribution, such as the upper tail slowly drifting from the 95th percentile to the 99.9th percentile. It enables the quantitative extraction of early abnormal features under strong noise and variable operating conditions, providing a reliable basis for subsequent screening of suspicious candidates and cross-sectional comparison.

[0029] In another preferred embodiment of the present invention, the transverse anomaly coefficient corresponding to the suspected candidate is calculated based on the obtained root mean square vibration. Specifically, the following steps are included: The obtained root mean square (RMS) vibrations are sorted in ascending order, and the ascending rank r of the RMS vibration X corresponding to the suspected candidate is obtained. The transverse anomaly coefficient is then calculated. N represents the total number of elements in the sorting process. This represents the median of the root mean square of the vibration.

[0030] It should be noted that by sorting the root mean square of vibration among compressors of the same model and operating conditions in the same workshop, the system uses ascending ranking and median to distinguish between normal and abnormally high equipment. When the vibration of a suspected candidate is less than the median, the coefficient is zero; when it is greater than or equal to the median, the coefficient increases linearly with the ranking. This quantifies the relative deviation of the candidate within the group, effectively eliminating environmental noise and individual equipment differences, and accurately identifying truly abnormal equipment.

[0031] In another preferred embodiment of the present invention, the number of compressors that are the same as the suspected candidate models and have the same current operating condition combination is counted. When the number is less than the preset judgment number, subsequent operations are stopped and no fault time prediction is performed.

[0032] Understandably, it's crucial to ensure the statistical reliability and engineering applicability of the cross-sectional comparison results. When the number of compressors with the same model and operating conditions as the suspected candidate is too small, the cross-sectional anomaly coefficients based on ascending ranking and median will lose representativeness due to insufficient sample size. The ranking may fluctuate drastically due to one or two extreme values, and the median may fail to reflect the normal population distribution. Forcing fault prediction under these circumstances can easily lead to false alarms or missed alarms. Therefore, by pre-setting the judgment quantity and stopping subsequent operations, the decision-making risks caused by small samples can be effectively avoided, preventing unreliable warnings under statistically invalid conditions.

[0033] In another preferred embodiment of the present invention, if the total working time of the compressor recorded as a real abnormality is less than or equal to the preset aging working time, and the number of failures is less than or equal to 3, then no compressor failure warning will be issued.

[0034] It is worth noting the importance of distinguishing between early, sporadic failures and the regular degradation that occurs during equipment aging. When the total operating time of the compressor is less than or equal to the preset aging operating time and the number of failures does not exceed three, the equipment is still in its initial operation or break-in period. At this time, failures are mostly caused by installation and commissioning, accidental disturbances, or early quality fluctuations, and the failure interval distribution is extremely unstable and lacks statistical regularity. If failure timing is forcibly predicted based on the mean and standard deviation of failure intervals at this stage, it is very easy to produce serious biases due to insufficient samples, leading to frequent false alarms. By skipping the failure warnings for this part of the equipment, unreliable alerts can be effectively avoided during the data statistical failure stage, maintaining the trust of maintenance personnel. After the equipment has accumulated a sufficiently long operating time and failure samples, prediction logic based on ideal failure cycles can be used, thereby achieving a reasonable balance between reducing the false alarm rate and ensuring timely warnings.

[0035] In another preferred embodiment of the present invention, if the sampling window includes a period of compressor shutdown, the data obtained from that sampling window is discarded.

[0036] It is worth noting that invalid or abnormal data from compressor shutdown periods should be excluded. Vibration, speed, and load during shutdown do not reflect the normal operating status of the equipment. Including these data in the statistical sample would lower the historical distribution of the root mean square of vibration, causing a baseline shift in normal operating conditions. This would interfere with the accuracy of quantile calculations and anomaly coefficients, thus reducing the risk of false alarms.

[0037] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A smart enterprise management platform based on big data, characterized in that, Includes the following modules: Window sampling module: Preset sampling windows, and after each sampling window is completed, obtain the root mean square vibration of the compressor, the average rotor speed V, and the average load F within that window; construct the operating condition combination of the compressor in that sampling window based on the average rotor speed V and the average load F, and record all sampling windows with the same operating condition combination as the same operating condition. Suspicious candidate modules: For each compressor, obtain the root mean square of vibration under the same operating conditions in sequence, and construct a list of root mean square vibration under the same operating conditions [x1, x2, ..., xi], where xi represents the root mean square of vibration in the i-th sampling window under the same operating conditions; Obtain the root mean square (RMS) X of the current sampling window and the corresponding list of RMS vibrations, and obtain the λ quantiles in the RMS vibration list. and γ quantile , where λ represents the preset first coefficient and γ represents the preset second coefficient; Based on λ quantiles and γ quantile Calculate the anomaly coefficient S of the compressor in the current sampling window; Horizontal comparison module: If S is greater than or equal to the preset screening threshold The compressor is designated as a suspected candidate. Within the same workshop, compressors of the same model and with the same current operating condition are selected. The root mean square (RMS) vibration of these compressors within the most recent complete sampling window is obtained. Based on the obtained RMS vibration data, the transverse anomaly coefficient corresponding to the suspected candidate is calculated. ; Fault prediction module: If the horizontal anomaly coefficient If the value is greater than or equal to the preset threshold for judging true anomalies, the corresponding suspicious candidate is recorded as a true anomaly. The mean T and standard deviation s of the historical fault intervals of the true anomaly are obtained, the ideal fault period Ts = T - 2s is calculated, and the time of the most recent fault of the true anomaly is obtained. When at time Send compressor fault warnings in real time.

2. The intelligent enterprise management platform based on big data according to claim 1, characterized in that, In the aforementioned window sampling module, constructing the compressor's operating condition combination within the sampling window based on the average rotor speed V and average load F specifically includes the following steps: Calculate the RPM gear and load settings Where v represents the preset basic speed and f represents the preset basic load, the operating condition combination of the compressor in this sampling window is generated. .

3. The intelligent enterprise management platform based on big data according to claim 1, characterized in that, In the aforementioned suspicious candidate modules, the first coefficient λ satisfies 90%≤λ≤95%, and the second coefficient γ satisfies γ≥99.9%.

4. The intelligent enterprise management platform based on big data according to claim 1, characterized in that, In the aforementioned suspected candidate module, based on λ quantile and γ quantile The method for calculating the anomaly coefficient S of the compressor in the current sampling window specifically includes the following steps: Calculate the anomaly coefficient of the compressor within the current sampling window. ,in, This represents the preset judgment threshold.

5. The intelligent enterprise management platform based on big data according to claim 1, characterized in that, In the aforementioned lateral comparison module, the lateral anomaly coefficients corresponding to the suspected candidates are calculated based on the acquired root mean square vibration. Specifically, the following steps are included: The obtained root mean square (RMS) vibrations are sorted in ascending order, and the ascending rank r of the RMS vibration X corresponding to the suspected candidate is obtained. The transverse anomaly coefficient is then calculated. N represents the total number of elements in the sorting process. This represents the median of the root mean square of the vibration.

6. The intelligent enterprise management platform based on big data according to claim 1, characterized in that, In the horizontal comparison module, the number of compressors that are the same as the suspected candidate models and have the same current operating condition combination is counted. When the number is less than the preset judgment number, subsequent operations are stopped and no fault time prediction is performed.

7. The intelligent enterprise management platform based on big data according to claim 1, characterized in that, In the fault prediction module, if the total working time of the compressor recorded as a real abnormality is less than or equal to the preset aging working time, and the number of faults is less than or equal to 3, then no compressor fault warning will be issued.

8. The intelligent enterprise management platform based on big data according to claim 1, characterized in that, In the aforementioned window sampling module, if the sampling window contains a period of compressor shutdown, the data obtained from that sampling window will be discarded.

Images