Prediction method and system based on industrial big data

Abstract

The invention relates to the field of big data analysis, in particular to a prediction method and system based on industrial big data. The method comprises the following steps: extracting an equipment data stream in a factory production process, and carrying out multi-cycle division and centralized visualization to generate a visual production data stream; performing deviation evaluation according to the visual production data flow, and marking potential abnormal equipment; calculating the change rate and the change amplitude of the potential abnormal equipment; positioning fault equipment according to the change rate and the change amplitude, and marking the fault equipment; performing multi-parameter linkage analysis and deviation attribution analysis on the fault equipment, and identifying equipment fault factors; and performing equipment failure shutdown prediction and equipment maintenance priority planning based on the equipment failure factors, and generating an equipment maintenance sequence. Based on factory big data analysis, the equipment fault time point is accurately predicted, and the accuracy of fault diagnosis and the shutdown maintenance efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, and more particularly to a prediction method and system based on industrial big data. Background Technology

[0002] With the continuous development of intelligent manufacturing and the Industrial Internet, equipment management and maintenance in factories are facing increasingly complex challenges. Traditional methods for predicting and repairing equipment failures often rely on human experience, regular inspections, and preventative maintenance based on simple indicators. These methods have significant limitations when dealing with rapidly changing production environments and complex equipment systems. As production scales up and equipment types diversify, the downtime and repair costs caused by equipment failures have a more significant impact on factory production efficiency and economic benefits. Therefore, improving the accuracy of equipment failure diagnosis, reducing repair costs, and minimizing equipment downtime have become critical issues that the manufacturing industry urgently needs to address.

[0003] Compared to traditional monitoring methods, modern factory equipment management increasingly relies on the integration of big data, artificial intelligence, and the Internet of Things (IoT) technologies. Especially driven by Industry 4.0 and smart factories, equipment fault diagnosis and maintenance decisions are moving towards greater intelligence and automation. Industrial big data provides a rich source of information for fault analysis and prediction by collecting multi-dimensional data such as equipment operating data, environmental information, and operation records. However, relying solely on data analysis often fails to fully reflect the health status of equipment, especially for issues such as external physical damage and surface anomalies, which traditional sensor data often struggles to detect. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a prediction method and system based on industrial big data, thereby resolving at least one of the aforementioned technical issues.

[0005] To achieve the above objectives, this invention provides a prediction method based on industrial big data, comprising the following steps: Step S1: Extract the equipment data stream of the factory production process, divide it into multiple cycles and centrally visualize it to generate a visualized production data stream; Step S2: Perform deviation assessment based on the visualized production data stream and mark potentially abnormal equipment; Step S3: Calculate the rate and magnitude of change of the potentially abnormal equipment; locate the faulty equipment based on the rate and magnitude of change, and mark the faulty equipment; Step S4: Perform multi-parameter linkage analysis and deviation attribution analysis on the faulty equipment to identify the factors causing the equipment failure; Step S5: Based on equipment failure factors, predict equipment failure downtime and plan equipment maintenance priorities to generate an equipment maintenance sequence.

[0006] This specification provides a prediction system based on industrial big data for executing the prediction method based on industrial big data as described above, including: The cycle segmentation module is used to extract equipment data streams from the factory production process, perform multi-cycle segmentation and centralized visualization, and generate a visualized production data stream. The deviation assessment module is used to assess deviations based on the visualized production data stream and mark potentially abnormal equipment. The fault location module is used to calculate the rate and magnitude of change of the potentially abnormal equipment; locate the faulty equipment based on the rate and magnitude of change; and mark the faulty equipment. The attribution analysis module is used to perform multi-parameter linkage analysis and deviation attribution analysis on faulty equipment to identify the factors causing equipment failure. The downtime prediction module is used to predict equipment failure downtime and plan equipment maintenance priorities based on equipment failure factors, and generate equipment maintenance sequences.

[0007] The beneficial effects of this invention are specifically as follows: By extracting production data streams (such as temperature, pressure, vibration, etc.) from factory equipment, the equipment status can be monitored in real time and comprehensively. This real-time monitoring helps identify potential problems in the production process and avoids equipment failures from causing widespread impact on production. Dividing and centrally displaying multi-period data helps operators or managers quickly identify equipment operating trends and determine whether equipment is malfunctioning. The graphical presentation makes data analysis more intuitive and facilitates decision-making. Periodic division allows viewing equipment data performance over different time periods, facilitating historical equipment analysis and identifying periodic problems or potential failures. Real-time deviation assessment of production data streams can quickly detect deviations from normal equipment conditions. Deviations may indicate equipment wear, signs of malfunction, or changes in the operating environment. Timely detection of these deviations helps prevent equipment failures. The deviation assessment system automatically marks potentially abnormal equipment, reducing the workload of manual monitoring and improving the accuracy and timeliness of fault warnings. By marking potentially abnormal equipment, production managers can intervene before problems escalate into failures, reducing the probability of equipment failures and thus avoiding production interruptions. Calculating the rate and magnitude of change in equipment data streams helps determine equipment failure trends and predict the timing of failures in advance. This rate of change analysis can reveal instabilities in equipment operation, allowing for proactive measures. By calculating the rate and magnitude of change, faulty equipment can be located more accurately, enabling rapid response and preventing the entire production line or process from halting due to equipment failure. Accurately tagged faulty equipment provides a basis for subsequent maintenance resource scheduling, ensuring maintenance personnel can quickly address the faulty equipment without wasting time on unnecessary troubleshooting. Multi-parameter linkage analysis comprehensively considers various equipment operating indicators, identifying multiple potential factors affecting equipment operation. This analytical method can more comprehensively identify the causes of failures, not just limited to a single parameter. Deviation attribution analysis allows for a better understanding of the root causes of equipment failures, leading to the development of targeted preventative measures. This helps prevent the recurrence of similar problems in the future, reducing repetitive failures. In-depth analysis of the relationships between multiple related parameters improves the accuracy of fault diagnosis, avoids misjudgments, and reduces equipment damage. Analysis of equipment failure factors can predict when equipment may fail or stop, allowing for proactive preparation. This predictive maintenance reduces equipment downtime and improves production efficiency. Based on equipment failure prediction, importance, and impact on production, a reasonable maintenance priority should be established. This not only helps to concentrate resources on repairing critical equipment but also avoids production bottlenecks caused by excessively long maintenance times or improper scheduling. Generating equipment maintenance sequences ensures that maintenance work is carried out efficiently according to priority. It avoids chaotic equipment maintenance work, ensures optimal allocation of equipment maintenance resources, and improves the overall operating efficiency of the production system. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the steps of a prediction method based on industrial big data according to the present invention; Figure 2 This is a detailed flowchart illustrating the implementation steps of step S1. Figure 3 This is a flowchart illustrating the detailed implementation steps of step S2. Detailed Implementation

[0009] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0010] This application provides a prediction method and system based on industrial big data. The execution entities of the prediction method and system based on industrial big data include, but are not limited to, the following: mechanical equipment, data processing platform, cloud server node, network upload device, etc., which can be considered as general computing nodes of this application. The data processing platform includes, but is not limited to, at least one of the following: audio and image management system, information management system, and cloud data management system.

[0011] Please see Figures 1 to 3 This invention provides a prediction method based on industrial big data, comprising the following steps: Step S1: Extract the equipment data stream of the factory production process, divide it into multiple cycles and centrally visualize it to generate a visualized production data stream; Step S2: Perform deviation assessment based on the visualized production data stream and mark potentially abnormal equipment; Step S3: Calculate the rate and magnitude of change of the potentially abnormal equipment; locate the faulty equipment based on the rate and magnitude of change, and mark the faulty equipment; Step S4: Perform multi-parameter linkage analysis and deviation attribution analysis on the faulty equipment to identify the factors causing the equipment failure; Step S5: Based on equipment failure factors, predict equipment failure downtime and plan equipment maintenance priorities to generate an equipment maintenance sequence.

[0012] In the embodiments of the present invention, see Figure 1 The diagram below illustrates the steps of a prediction method based on industrial big data according to the present invention. In this example, the steps of the prediction method based on industrial big data include: Step S1: Extract the equipment data stream of the factory production process, divide it into multiple cycles and centrally visualize it to generate a visualized production data stream; In this embodiment, various types of data generated during equipment operation are extracted from multiple data sources, including the factory's industrial control system (SCADA system), programmable logic controller (PLC), human-machine interface (HMI), and sensor acquisition network. This data includes: real-time operating parameters (equipment speed, pressure, temperature, flow rate, power, etc.), energy consumption data (current, voltage, active power, reactive power, etc.), vibration acceleration data (three-axis acceleration values), position encoder feedback data, production output data, process parameter settings, operator instructions, etc. The data acquisition time span should cover a historical period of 3-6 months to ensure a sufficient number of complete production cycle samples.

[0013] Data is divided according to the factory's actual production plan cycle. Typical division methods include: daily cycle division (using 24-hour data as a basic cycle unit), weekly cycle division (using 7-day cycles to consider differences in production rhythm across different workdays), and monthly cycle division (using a calendar month or accounting month as the cycle). Taking the daily cycle as an example, if a production line is designed to operate on a 24-hour cycle, starting at 8:00 AM the next day and ending at 8:00 AM, then the data for each 24-hour period is divided according to this boundary, forming independent cycle datasets. For multi-shift production (e.g., three-shift operation), it may be necessary to divide the data by shift, with each shift consisting of 8 hours as a sub-cycle.

[0014] Time synchronization and alignment are performed on the extracted data from each period. In multi-source heterogeneous data acquisition, the sampling frequencies of different sensors may vary significantly—for example, a vibration sensor may sample at 1000Hz, while a temperature sensor may sample every 10 seconds. A unified timestamp reference system (based on the GPS clock of the central control system) needs to be established to map all data onto a unified time axis. For cases with inconsistent sampling frequencies, a time window aggregation method is used: high-frequency data (vibration) is downsampled and feature extracted. Within each time window (e.g., 1 minute), statistical characteristics such as mean, variance, peak value, peak-to-peak value, and frequency domain energy are calculated. Multiple high-frequency sampling points are compressed into a single feature value, which is then fused with low-frequency data.

[0015] The aligned multi-cycle data is visualized in a hierarchical manner. A multi-dimensional data flow visualization is established: First, a time-series chart of a single device is plotted, with the horizontal axis representing time (hours) and the vertical axis representing parameter values. Curves showing the changes of key parameters (temperature, vibration, current, etc.) throughout the complete production cycle are drawn. Taking temperature as an example, under normal circumstances, a smooth curve should be displayed, rising during the cold start phase → experiencing limited fluctuations during the stable operation phase → declining during the shutdown cooling phase, forming a typical "mountain-shaped" curve. Second, data from multiple identical cycles are overlaid to create a "cycle overlay chart." For example, daily cycle data from the past 30 days is selected, and the transparency of the 30 daily cycle curves is set to 10%, all overlaid on the same coordinate system. If production is stable, these 30 curves should highly overlap, forming a clear "mean trajectory." If the curve of a particular day deviates significantly from the other curves, an abnormal cycle can be intuitively identified.

[0016] Further develop a data flow visualization for the equipment group. Create a heatmap representation where the horizontal axis represents the numbers of all production equipment within the factory, and the vertical axis represents time (arranged by cycle). Each cell is color-coded to represent the average value of a key parameter for that equipment within that cycle. For example, a heatmap can show the average current value of all compressors over the past 30 production cycles: normally it should be a uniform light blue; if a compressor shows a dark red color in a particular cycle (indicating abnormally high current), the anomaly can be quickly identified. Simultaneously, generate an equipment interconnection diagram, using lines to represent the production coupling relationships between equipment and marking the data flow direction, facilitating subsequent tracing of the problem's source.

[0017] Statistical features are extracted based on visualization. Multidimensional statistical indicators are calculated for each period's data: mean (reflecting the central location of the parameter), standard deviation (reflecting the amplitude of fluctuation), maximum value, minimum value, interquartile range (IQR, reflecting the fluctuation range of the middle 50% of the data), skewness (reflecting the asymmetry of the data distribution, which may indicate abnormal operating conditions), and kurtosis (reflecting the sharpness of the distribution; high kurtosis may indicate the presence of peak anomalies), etc. For energy consumption parameters, total energy consumption and energy efficiency (output / energy consumption) for the period also need to be calculated. These statistical features form a "period feature vector," with one feature vector representing each complete period, establishing a quantitative basis for the next step of deviation assessment.

[0018] All visualization results are centralized on a unified data display platform, enabling production managers to clearly observe the overall production status of the factory. This platform should have interactive functions, supporting operations such as filtering by equipment type, querying by time range, and zooming in and out to view details. Through such centralized visualization, a "standard data flow image" of the factory's production process is established, laying the foundation for deviation identification and anomaly localization in subsequent steps.

[0019] Step S2: Perform deviation assessment based on the visualized production data stream and mark potentially abnormal equipment; In this embodiment, a multi-dimensional evaluation system for deviation calculation is established. For each device, the deviation from the historical baseline is calculated for each parameter in each cycle. Specific methods for deviation calculation include: absolute deviation (current cycle parameter value - historical cycle average), relative deviation (absolute deviation / historical cycle average × 100%), standardized deviation (absolute deviation / historical cycle standard deviation, also known as Z-score), percentile deviation (the percentile ranking of the current value relative to the historical data distribution), and other dimensions. Among these, standardized deviation (Z-score) is the most important indicator, reflecting the distance between the current value and the historical average. Using standard deviation as the unit, it makes the deviations of different parameters comparable. For example, the normal fluctuation range of a temperature sensor is 68-72℃, with a standard deviation of 1.2℃; while the normal fluctuation range of a pressure sensor is 4.8-5.2 bar, with a standard deviation of 0.15 bar. If the current temperature is 75℃, then Z-score = (75-70) / 1.2 = 4.17; if the current pressure is 5.5 bar, then Z-score = (5.5-5.0) / 0.15 = 3.33. Using a unified Z-score, the degree of deviation between the two parameters can be directly compared.

[0020] A hierarchical deviation threshold system is defined. Based on the general patterns of fault evolution and industry experience, three progressive thresholds are set: the first-level threshold (micro-deviation threshold) is Z-score > 1.5, indicating that the parameter deviates from the historical average by 1.5 standard deviations, which is a slight anomaly, on the edge of normal fluctuation; the second-level threshold (meso-deviation threshold) is Z-score > 2.5, indicating a deviation of 2.5 standard deviations, which is a significant anomaly requiring attention; the third-level threshold (macro-deviation threshold) is Z-score > 3.5, indicating a deviation of 3.5 standard deviations, which is a severe anomaly with significant risk. These three thresholds are set based on statistical principles: under a normal distribution, 1.5 standard deviations corresponds to approximately 93% of the data falling within the range, 2.5 corresponds to approximately 98.8%, and 3.5 corresponds to over 99.95%. Therefore, data with a Z-score exceeding 3.5 has an extremely low probability of occurrence and almost certainly represents signs of a fault.

[0021] Real-time deviation monitoring and evaluation are performed. When new production cycle data enters the system, the deviation index for each parameter of each piece of equipment is calculated individually. A "deviation evaluation framework" is established: for each monitored parameter of each piece of equipment, the current value is compared with the historical baseline to identify whether a certain threshold has been exceeded. Taking a major production piece of equipment (such as an injection molding machine) as an example, its key monitored parameters include: barrel temperature (target value 200℃, allowable deviation ±5℃), injection pressure (target value 100bar, allowable deviation ±10bar), screw speed (target value 50rpm, allowable deviation ±5rpm), hydraulic oil temperature (target value 50℃, allowable deviation ±3℃), etc. Assuming that the average barrel temperature over the past 30 cycles is 200.2℃ and the standard deviation is 1.8℃; if the barrel temperature in the current cycle is 206℃, then the Z-score = (206-200.2) / 1.8 = 3.22, exceeding the mesoscopic threshold of 2.5, which is a significant anomaly.

[0022] Establish a system for identifying "parameter anomaly combination patterns." A slight exceedance of a single parameter may be normal fluctuation, but the simultaneous anomaly of multiple related parameters is more likely to indicate a real fault. Establish a physical correlation matrix between parameters: for example, in a hydraulic system, if the pump's outlet pressure increases, the hydraulic oil temperature will inevitably increase; if the pressure increases but the temperature remains unchanged, it may indicate a temperature sensor malfunction or a cooling system abnormality. Therefore, when a main parameter anomaly is detected, cross-check whether its related parameters are also synchronously anomaly-free. Use the "parameter anomaly coupling degree" index to quantify this correlation: if multiple parameters have similar anomaly threshold levels within the same period and all exceed the first-level threshold, the anomaly coupling degree for that period is high, marked as "multi-parameter linkage anomaly"; otherwise, it is "single-parameter isolated anomaly," which has a lower probability of being genuine and may be sensor noise or temporary interference.

[0023] Mark and classify potentially abnormal equipment. Based on the above deviation assessment results, calculate an "abnormal risk score" for each device. The scoring rules are as follows: if the device has a parameter with a Z-score > 3.5 in the current period, add 50 points; for a parameter with a Z-score > 2.5, add 30 points; for a parameter with a Z-score > 1.5, add 10 points; if there are multiple parameter linkage anomalies (3 or more parameters are abnormal at the same time), add an additional 20 points on top of the above scores. Based on the score, the device is divided into three levels: score 0-20 is green (normal), score 21-50 is yellow (minor anomaly, requires monitoring), and score > 50 is red (obvious anomaly, requires immediate attention). For example, if an air compressor exhibits abnormalities in three parameters—exhaust temperature (Z-score=3.8), current (Z-score=2.9), and vibration (Z-score=2.1)—during a certain cycle, and there is a clear correlation between these parameters (increased pressure → increased current → increased temperature), then the abnormality score for that cycle is 50 + 30 + 10 + 20 = 110 points, and it is marked as a red abnormal device.

[0024] Perform repeatability checks for abnormal cycles. To avoid false alarms caused by random fluctuations in a single cycle, establish an "abnormal persistence check" mechanism: if a device is marked as yellow or red abnormal for two consecutive cycles, it is confirmed as a "real abnormality"; if an abnormality only occurs in a single cycle, it is marked as a "temporary abnormality" and monitoring continues. The reason for this is that real equipment failures usually show abnormal signs continuously in subsequent cycles, while abnormalities caused by sensor noise or temporary interference are usually one-off events.

[0025] All marked potentially abnormal devices are compiled into an "Abnormal Device Monitoring List," which records information such as the name of the abnormal parameter, the degree of abnormality score, the time of the first occurrence of the abnormality, and the number of periods of continuous abnormality for each device. This list lays the foundation for the next step of rate amplitude analysis.

[0026] Step S3: Calculate the rate and magnitude of change of the potentially abnormal equipment; locate the faulty equipment based on the rate and magnitude of change, and mark the faulty equipment; In this embodiment, the potential abnormal equipment marked in S2 is dynamically tracked over time. An "abnormal parameter monitoring sequence" is established, and for each equipment marked as abnormal, the Z-score values ​​of its various deviation parameters are recorded for each production cycle. For example, for an air compressor, the Z-score values ​​of its exhaust temperature over 10 consecutive cycles from the discovery of the abnormality may present the following sequence: [1.8, 2.2, 2.7, 3.1, 3.5, 3.8, 4.0, 4.1, 4.0, 3.9]. This sequence clearly shows the evolution of the fault. The first derivative (rate of change) and second derivative (acceleration of change) of the deviation are calculated. The first derivative reflects the amount of change of the parameter deviation per unit time, mathematically represented by the difference in Z-score between two adjacent cycles. Taking the temperature Z-score sequence above as an example, the first derivative sequence is: [0.4, 0.5, 0.4, 0.4, 0.3, 0.2, 0.1, -0.1, -0.1], with units of "Z-score / cycle". This indicates that the deviation increases at an average rate of 0.38 Z-score / cycle within the first 5 cycles, after which the growth rate slows down. The second derivative reflects the change in the rate of change itself, i.e., the acceleration of change: [-0.1, -0.1, 0, 0, -0.1, -0.1, -0.2, 0, 0]. Negative values ​​indicate that the growth rate is slowing down. The "fault evolution stage" is defined based on the rate of change. The evolution of a fault is divided into different stages, each corresponding to a specific rate of change range: early nascent stage (rate of change 0.1-0.2 Z-score / cycle), accelerated evolution stage (rate of change 0.2-0.5 Z-score / cycle), rapid deterioration stage (rate of change > 0.5 Z-score / cycle), and stabilization or mitigation stage (rate of change < 0.1 Z-score / cycle or negative). These stage divisions are based on statistical analysis of typical fault cases: Analysis of 100 historical equipment fault cases revealed that when the rate of increase in parameter deviation exceeds 0.5 Z-score / cycle, the fault typically erupts within 2-4 cycles; when the rate of increase is in the range of 0.2-0.5, there is a longer available warning time (usually 4-8 cycles); a rate of increase below 0.2 indicates slow fault evolution, with possibly 1-2 months for preventative maintenance. The "cumulative deviation magnitude of fault evolution" is calculated. This is the total change in parameter deviation from the initial appearance of the fault to the current moment. The calculation method is to subtract the historical baseline Z-score of the parameter (usually 0) from the current period's Z-score value. Taking temperature as an example, the cumulative increase from Z-score = 1.8 (the first abnormal period) to Z-score = 4.1 (currently) is 4.1 - 0 = 4.1 standard deviations.This indicator reflects the extent to which the fault has accumulated. Combined with the rate of change, it can predict when the fault will reach a critical value. A comprehensive evaluation model of "multi-parameter amplitude and rate" needs to be established. In actual faults, multiple parameters often become abnormal simultaneously. A comprehensive fault severity scoring model needs to be established. The scoring rule is as follows: calculate the average Z-score value of all abnormal parameters (Z-score > 1.5) within the current period, denoted as "average deviation degree"; calculate the average rate of change of these abnormal parameters, denoted as "average evolution rate"; comprehensive score = average deviation degree × 40% + average evolution rate × 60%. The weighting reflects that the rate of change is more important than the deviation degree itself, because rapidly deteriorating small faults are more risky than slowly progressing large faults. For example, if the abnormal parameters of a certain device are evaluated as: average deviation degree = 3.2, average evolution rate = 0.35 Z-score / period, then the comprehensive score = 3.2 × 0.4 + 0.35 × 0.6 = 1.28 + 0.21 = 1.49. The severity of equipment malfunctions is graded based on a comprehensive score. Grading thresholds are set as follows: a comprehensive score of 0-0.5 indicates "potentially malfunctioning equipment" (in the early monitoring stage); 0.5-1.0 indicates "minor malfunctioning equipment" (requiring immediate maintenance); 1.0-1.5 indicates "moderate malfunctioning equipment" (requiring immediate maintenance); and >1.5 indicates "severe malfunctioning equipment" (requiring emergency shutdown for maintenance). In a production cycle, the system might identify: Air compressor A with a comprehensive score of 1.32 (moderate malfunction), coolant pump B with a score of 0.58 (minor malfunction), injection molding machine C with a score of 0.35 (potentially malfunctioning), and hydraulic station D with a score of 1.72 (severe malfunction). Inflection point identification is then performed. In some cases, the rate of change of the malfunction may abruptly change, accelerating suddenly from a slow increase or reversing abruptly from acceleration (possibly due to temporary adjustments by operators). These critical inflection points can be identified by recognizing the sign reversal points or extreme points of the first derivative of the Z-score sequence. For example, in the temperature sequence above, the first derivative suddenly changes from 0.1 to -0.1 in the 8th cycle, indicating that the temperature has started to drop. This could be because maintenance personnel have partially repaired the fault after discovering it, or the system has automatically adjusted its temperature control. These inflection points are important because they may indicate a change in the direction of the fault evolution. The "rate of change" and "amplitude of change" are used in combination for the final location and labeling of the faulty equipment. Based on all the aforementioned analysis results, the system generates a "List of Faulty Equipment," recording for each faulty device: initial fault type assessment (inferred from combinations of abnormal parameters, such as "cooling system anomaly"), current severity level, Z-score of current deviation, average evolution rate, expected number of cycles for the fault to worsen, and whether an emergency shutdown is required.For example, the record for hydraulic station D might be: "Faulty equipment number: D1, Fault type: abnormal hydraulic pump pressure, Severity: severe (score 1.72), Current deviation Z-score: 4.3, Average evolution rate: 0.52 Z-score / cycle, Expected deterioration time: 2-3 cycles, Recommendation: Immediately conduct maintenance assessment".

[0027] Step S4: Perform multi-parameter linkage analysis and deviation attribution analysis on the faulty equipment to identify the factors causing the equipment failure; In this embodiment, a "physical parameter correlation model" for the target device is established. Different types of equipment have different physical laws. Taking an air compressor as an example, a causal relationship network of its key parameters is established: the intake air volume is affected by the intake valve state → the compression process leads to increased pressure → increased pressure leads to increased current → the heat generated by compression work leads to increased exhaust temperature → increased temperature triggers the cooling system → the cooling effect affects oil temperature. Therefore, if the intake filter is clogged, it will lead to: reduced intake air volume → increased compression pressure (because the mass of the same volume of gas increases) → increased current → further increased temperature → increased cooling system load. This forms a causal chain of "increased intake resistance → abnormal pressure → abnormal current → abnormal temperature".

[0028] Perform "time series benchmarking analysis of parameter anomalies." Display the time series curves of multiple key anomaly parameters in parallel, observing their phase relationships and time lags. For example, in a hydraulic system fault, observe the anomaly sequences of pump outlet pressure, pump motor current, hydraulic oil temperature, and cooling water flow rate. A normal causal relationship should be: within 1-2 cycles after the pressure increases, the current also increases; within 2-4 cycles after the current increases, the oil temperature increases; and within 3-5 cycles after the oil temperature increases, the cooling water flow rate should increase. Observing such a reasonable time series relationship verifies the rationality of the fault propagation chain. However, if an increase in oil temperature is observed but no increase in cooling water flow rate, it may indicate a fault in the cooling system itself (such as a cooling pump failure or pipe blockage).

[0029] Perform a "correlation analysis between parameters." Use the Pearson correlation coefficient to calculate the degree of linear correlation between multiple outlier parameters. The correlation coefficient ranges from -1 to +1, with a larger absolute value indicating a stronger correlation. For example, calculating the correlation coefficient between the exhaust temperature and exhaust pressure of an air compressor over the past 20 periods, assuming a value of 0.89, indicates a strong positive correlation, consistent with the physical law (higher pressure → higher temperature). However, if the correlation coefficient between exhaust temperature and intake air flow is calculated to be 0.92 (strong positive correlation), this violates common sense (less intake air → higher compression ratio → higher temperature should be stronger), requiring further investigation to determine if other influencing factors exist. Correlation analysis helps identify which parameter associations are expected and which are outliers.

[0030] Perform "fault feature pattern matching". Extract fault cases with similar characteristics from the historical fault case database to establish a "fault feature database". For example, for air compressors, establish the following feature database entries: Feature Pattern 1 (intake filter blockage): Features include decreased intake pressure, increased exhaust pressure, increased current, rapid increase in oil temperature, and decreased air production; Feature Pattern 2 (cooling system failure): Features include continuously increased oil temperature, abnormal cooling water flow or temperature, and basically normal pressure; Feature Pattern 3 (exhaust valve leakage): Features include pressure not rising, current increasing but less than normal, and abnormally high exhaust temperature; Feature Pattern 4 (motor failure): Features include severe current fluctuations or three-phase imbalance, and correspondingly increased pressure and temperature fluctuations. Match the current faulty equipment's multi-parameter abnormal combinations with these feature patterns and calculate the similarity score. The similarity calculation method is as follows: For each feature pattern, count how many of the current equipment's abnormal parameters overlap with that pattern, calculate the number of matching parameters / the total number of parameters in the pattern, and adjust using weights (key parameters have higher weights than secondary parameters). Assuming the current abnormal parameter combination of the air compressor is "decreased intake pressure (weight 1.0), increased exhaust pressure (weight 1.0), increased current (weight 1.0), increased oil temperature (weight 0.9), and decreased air production (weight 0.8)," the total weight sum is 4.7. The weight sum of feature pattern 1 is also 4.7. Therefore, the similarity score = (actual matching parameter weight sum / pattern weight sum) × 100%. Assuming a match of 4.5 / 4.7 = 95.7%, the similarity between this fault and pattern 1 is 95.7%.

[0031] Perform "parameter anomaly tracing." Trace the causal chain between parameters upwards to find the "source of the anomaly parameter." For example, if oil temperature, current, and pressure anomalies occur simultaneously, observe the time sequence. The anomaly first appears in the pressure parameter (in the first cycle), then the current parameter (in the second cycle), and finally the oil temperature parameter (in the fourth cycle). This indicates that the source of the anomaly is pressure. Further analyze the possible causes of the pressure anomaly—is it an intake-side problem (such as a clogged filter or a faulty intake valve) or an exhaust-side problem (such as a leaking exhaust valve or blocked pipes)? Further determine the cause by checking if the intake pressure drops synchronously. If the intake pressure is normal but the exhaust pressure increases, the problem lies at the exhaust end; if both pressures increase, the problem may be a decrease in compressor efficiency.

[0032] Perform "multi-dimensional cause hypothesis and verification". Based on the results of parameter linkage analysis, generate multiple possible failure cause hypotheses. For the air compressor example, the hypothesis set might include: Hypothesis 1 (intake filter blockage), Hypothesis 2 (cooling system failure), Hypothesis 3 (compressor wear), Hypothesis 4 (valve leakage), etc. Design "verification indicators" for each hypothesis—that is, what characteristics should be observed on which parameters if the hypothesis is true. For example, if the intake filter is blocked, the following should be observed: a continuous decrease in intake pressure, a decrease in intake flow, increased compressor power (increased current), and decreased efficiency (a decrease in the ratio of air production to current consumption). Test these indicators one by one in the current monitoring data and calculate the "verification fit degree" of each hypothesis. Hypothesis 1 has a fit degree of 87% in the current data (3.5 out of 4 verification indicators are confirmed), Hypothesis 2 has a fit degree of 42%, and Hypothesis 3 has a fit degree of 65%. Therefore, Hypothesis 1 is the most likely cause of failure.

[0033] Conduct a comprehensive analysis of the influencing factors. Collect all information related to the fault: equipment operating conditions (load rate, operating duration, whether it has been operating at high load for a long time), maintenance history (last maintenance time, maintenance content, post-maintenance operating time), environmental conditions (factory temperature, humidity, presence of vibration sources), operation records (whether there were any abnormal operations, whether parameter settings were changed), parts replacement records (whether critical parts were recently replaced), etc. Combining this information with abnormal parameters helps to form a more complete picture of the fault's cause. For example, if it is known that the air compressor has been operating at overload (exceeding 120% of the design air output) for the past two weeks before the anomaly was discovered, this strengthens the credibility of the "compressor wear" hypothesis, because long-term high-load operation accelerates mechanical wear, leading to increased clearances, increased leakage, and decreased efficiency.

[0034] Finally, a "Fault Factor Attribution Report" is generated. For each faulty device, the system generates a detailed cause analysis report, including: a description of the fault phenomenon (which parameters are abnormal and the degree of abnormality), an anomaly timeline (the order and time difference of the occurrence of each parameter anomaly), feature matching results (which cases are similar to those in the historical fault database), source anomaly identification (what is the root cause), influencing factor analysis (which operating conditions or maintenance history may have exacerbated the fault), and fault mechanism analysis (how the fault is generated and propagated at the physical level). For example, a hydraulic pump report might read: "Equipment No. HP-02, Fault Symptoms: Pressure increased by 3.8 Z-score, current abnormally increased by 2.9 Z-score, oil temperature rapidly increased by 3.5 Z-score. Abnormal Timeline: Pressure first exceeded the threshold in the first cycle, followed by current in the second cycle, and oil temperature showed a significant increase in the third cycle. Feature Matching: 89% similarity to the 'internal leakage expansion' pattern in the historical case database. Root Cause: The internal clearance of the hydraulic pump may have widened, leading to increased internal leakage, manifested as pressure failing to rise while oil temperature increases. Influencing Factors: This pump has been operating above 80% load for the past 3 months, accelerating wear. Recommendation: Immediately conduct an internal inspection of the pump; if clearances exceed specifications, repair or replacement should be performed." Step S5: Based on equipment failure factors, predict equipment failure downtime and plan equipment maintenance priorities to generate an equipment maintenance sequence.

[0035] In this embodiment, the "equipment shutdown threshold" is determined. Each type of equipment has a defined "unavailable state," that is, at what level certain key parameters of the equipment cannot continue its production tasks. This threshold is closely related to the equipment's functional requirements and process requirements. For example, for an air compressor, if its output pressure drops to a level that cannot drive downstream actuators (e.g., pneumatic grippers require at least 4.5 bar), or its output flow rate drops to a level that cannot meet the gas demand of the production line (the production line requires 15 m³ / min, while the compressor can only provide 10 m³ / min), then the air compressor is essentially ineffective. For an injection molding machine, if the barrel temperature deviates from the target value by more than ±8°C and cannot be restored by adjustment, or if the screw speed fluctuates by more than ±20%, it will affect product quality and may force a shutdown. For a hydraulic pump, if the output pressure is lower than 85% of the system's required operating pressure, or the flow rate drops by more than 20%, the hydraulic system cannot function properly. The system needs to clearly define its "shutdown threshold" for each piece of equipment, which is usually determined by the equipment manufacturer's technical specifications, production process requirements, industry standards, etc.

[0036] Perform "extrapolation prediction of fault evolution trajectory". Based on the parameter change rate and current deviation calculated in S3, a mathematical model is used to predict the future evolution of the parameters. For parameters that deteriorate linearly (with a basically constant change rate), a simple linear extrapolation can be used: Predicted value = Current value + Change rate × Number of prediction periods. For example, if the exhaust temperature of an air compressor has changed at a rate of 0.35℃ / cycle (Z-score dimension) over the past 5 cycles, the current temperature is 75℃, and the shutdown critical temperature is 85℃, then the extrapolation prediction requires (85-75) / 0.35 ≈ 28.6 cycles, meaning that it may automatically shut down due to excessive temperature in about 4 weeks. However, the evolution of many faults is non-linear, especially when approaching the critical value, where accelerated deterioration often occurs. For nonlinear deterioration parameters, more complex models are needed, such as using a historical similar case library for "trajectory matching prediction": This involves finding the most similar historical case (already matched in S4) from the case library, extracting the time span from the current stage to shutdown for that historical case, adjusting for differences in operating conditions, and obtaining a "reference prediction time". For example, if two historical cases with a similarity exceeding 85% are found in the case library, one taking 18 cycles to shutdown from the same fault stage and the other taking 24 cycles, and the current equipment's operating conditions are slightly worse than both historical cases (temperature rise is faster), then the predicted downtime for the equipment is (18+24) / 2 × 0.85 ≈ 17.9 cycles, or approximately 2-3 weeks.

[0037] A comprehensive prediction of shutdown timing under the combined effect of multiple parameters is needed. In reality, equipment shutdown is usually not caused by a single parameter reaching a threshold, but by the combined effect of multiple parameters. For example, a hydraulic system shutdown might be due to low pressure and insufficient flow, or excessively high oil temperature and reduced cooling efficiency. A "multi-parameter shutdown decision model" needs to be established. The most conservative approach is to use "parallel AND logic": the equipment will shut down as soon as any critical parameter reaches its shutdown threshold. Therefore, shutdown timing = min(predicted time for each critical parameter to reach its threshold). For example, if the pressure of a piece of equipment is expected to reach the critical value after 20 cycles, and the temperature is expected to reach the critical value after 15 cycles, then the shutdown timing of the equipment should be predicted based on 15 cycles. However, this method may be too conservative. A more realistic approach is to determine based on the equipment's protection devices and safety logic: which parameter anomalies will immediately trigger automatic protection shutdown, and which parameter anomalies allow for manual adjustment and temporary responses. After comprehensively considering these factors, the "most likely shutdown timing" is derived.

[0038] Conduct a "correlation analysis of the downtime of multiple faulty devices." In a factory's production system, the operation of certain devices is interdependent. For example, a compressor supplies gas to multiple gas-consuming devices; if the compressor stops, all gas-consuming devices may cease operation. A hydraulic station supplies hydraulic fluid to multiple hydraulic devices; if the hydraulic station stops, all these devices will stop. Based on predicting the downtime of individual devices, a "cascading downtime analysis" is needed. If the downtime of multiple critical devices is found to be concentrated in the near future (e.g., all within the next 2-3 weeks), the serious consequences of simultaneous downtime of these devices and the scarcity of maintenance resources should be considered. This will affect the final maintenance priority decision.

[0039] Conduct a "conflict analysis between production schedule and downtime". The factory has production schedules and order delivery dates. Some periods may be peak production periods, during which any equipment downtime will cause huge economic losses; while other periods may be windows for planned downtime maintenance, during which maintenance costs are relatively low. It is necessary to benchmark the predicted downtime points against the production schedule and analyze: "What will be the consequences if the failure happens during a peak production period?", "Are there planned downtime windows for maintenance that can be completed before the downtime occurs?", and "If maintenance is delayed, which order delivery dates will be delayed, and what will the potential economic losses be?" These analyses provide cost data for the next maintenance decision.

[0040] A multi-dimensional scoring system for maintenance priority is implemented. For each faulty piece of equipment, a maintenance priority score is calculated, taking into account multiple factors: the severity of the fault (the overall score in S3, weight 20%), the urgency of the downtime risk (the shorter the expected downtime, the higher the weight, weight 25%), the impact of the fault on production (if the downtime will affect multiple production lines or critical orders, the weight is high, weight 30%), the difficulty and time required for maintenance (simple and quick maintenance has a higher priority, weight 15%), and the availability of spare parts (if external spare parts need to be purchased, the priority is relatively lower, weight 10%). The specific calculation formula is: Maintenance Priority Score = Fault Severity × 0.2 + Downtime Urgency × 0.25 + Production Impact × 0.3 + Maintenance Operability × 0.15 + Spare Parts Availability × 0.1. For example, the system may calculate: Equipment A score = 8.5 (high priority), Equipment B score = 6.2 (medium priority), and Equipment C score = 4.1 (low priority).

[0041] Generate an "equipment maintenance sequence." Sort all faulty equipment according to their maintenance priority scores from highest to lowest, generating an ordered maintenance sequence list. However, this list cannot directly guide the actual maintenance work because "resource constraints" must also be considered. For example, if the maintenance team only has 2 people, but high-priority equipment requires 3 people to work simultaneously, or if the spare parts needed for some high-priority equipment are still being procured, then "resource feasibility verification" and "time scheduling optimization" are required. Using queuing theory or scheduling algorithms, under the given maintenance resource constraints, a specific "maintenance implementation plan" is formulated: specifying which equipment will be maintained at what time, which maintenance team will be responsible, the expected time, and the required spare parts, etc. For example: Monday morning, maintenance team A is responsible for equipment A (estimated 2 hours), and maintenance team B is responsible for equipment B (estimated 3.5 hours); Monday afternoon, after equipment A is maintained, team A moves on to preparations for equipment D; equipment D is maintained on Tuesday morning, and so on.

[0042] Conduct an "economic assessment of the maintenance plan." For each planned maintenance, calculate its costs and benefits. Costs include: maintenance labor costs (labor wages × man-hours), spare parts costs, and production loss costs due to equipment downtime (if maintenance requires equipment downtime). Benefits include: eliminating the risk of failure through maintenance, avoiding greater losses caused by sudden downtime, and potential order delay penalties. Use ROI (Return on Investment) or NPV (Net Present Value) to assess the economic rationale for the maintenance. For example, if the maintenance cost for equipment A is 2000 yuan (labor cost 1500 + spare parts cost 500), and the expected downtime loss avoided through maintenance is 15000 yuan (equipment downtime affects 4 hours of production, with a production loss of 3750 yuan per hour), then ROI = (15000-2000) / 2000 = 650%, indicating that the maintenance has high economic value. If the ROI of a maintenance is negative, it means that the maintenance cost exceeds the avoidable loss, and it may be necessary to reassess whether it is worthwhile.

[0043] Finally, information such as the maintenance sequence, implementation plan, resource allocation, and economic assessment is compiled into an "Equipment Maintenance Priority Planning Report" and delivered to the production and maintenance departments as guidance. The report clearly lists: the equipment to be maintained and its priority ranking, the expected downtime risk points, the recommended maintenance time windows, the required manpower and spare parts resources, the expected maintenance costs and benefits, and possible alternatives. This planning report will drive subsequent specific maintenance actions.

[0044] In this embodiment, see Figure 2 The diagram below illustrates the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: Extracting equipment data streams from the factory production process based on multiple device sensors; Determine the data sampling frequency of the multi-device sensors; calculate the average frequency based on the data sampling frequency to generate a standard sampling frequency; The device data stream is time-aligned according to the standard sampling frequency to generate a time-synchronized data stream. Sensor noise detection and adaptive filtering optimization are performed on the time-synchronized data stream to generate a filtered and noise-reduced data stream. Define a segmentation period, and divide the filtered and denoised data stream into multiple periods based on the segmentation period to generate multiple time window data streams; Centralized visualization of multi-device data streams across multiple time windows generates a visualized production data stream.

[0045] In this embodiment, during factory production, multiple key devices collect operational status information through sensors to form a device data stream. Sensor types include vibration sensors, temperature sensors, current sensors, oil pressure sensors, and acoustic monitoring modules, each measuring different physical quantities with significant differences in sampling range and sensitivity. For example, vibration sensors typically have a measurement range of 0 to 50 g and a sampling frequency of 1 kHz; temperature sensors have a measurement range of -20 to 150°C with an accuracy of approximately ±0.5°C; and current sensors have a sampling range of 0 to 100 A with a resolution of 0.01 A. Data acquisition is timestamped to ensure millisecond-level time accuracy. The data stream is temporarily stored in a local buffer during acquisition to prevent data loss due to momentary network latency. The acquired data is stored in the form of device number, sensor type, acquisition time, and data value, forming a continuous data sequence. This stage ensures that each data point contains complete spatiotemporal information, providing a basis for subsequent sampling frequency calculation and timing alignment, while also ensuring data integrity and consistency. Since the sampling frequencies of different device sensors differ, it is necessary to accurately determine the actual sampling rate of each sensor to achieve unified data processing. Vibration sensors typically use sampling frequencies of 1 kHz to 2 kHz, temperature and pressure sensors 10 Hz to 50 Hz, and current sensors around 500 Hz. The sampling frequency is determined using a time interval statistical method. The average sampling interval is calculated by determining the time difference between consecutive sampling points, and then converted into the sampling frequency. To avoid the impact of occasional data loss or clock drift, a sliding time window is used for frequency calculation on each data stream. The window length is typically 2 to 5 seconds, and the average value of multiple sub-intervals is calculated. For sensors with large frequency fluctuations, a maximum tolerance deviation of ±5% can be set. If the deviation exceeds the threshold, the sensor data acquisition status is reassessed. This method allows for the acquisition of stable sampling frequencies for each sensor, providing a reliable basis for subsequent standardization of sampling frequencies.

[0046] After obtaining the sampling frequencies of each sensor, a unified standard sampling frequency needs to be generated to align the data from multiple devices. The standard sampling frequency is determined based on the importance and stability of each sensor's frequency. For example, vibration and current sensors have a significant impact on the dynamic state of the equipment, and their sampling frequency weights can be set to 0.3 to 0.35; temperature and oil pressure sensors can have weights set to 0.05 to 0.1. By calculating a weighted average of each sampling frequency, a unified sampling interval is generated, typically between 50 and 100 milliseconds, ensuring that the data from most sensors can fully express their dynamic characteristics at this time step. The standard sampling frequency provides a time reference for subsequent data interpolation and time-series synchronization, while ensuring that the data from different sensors can directly correspond in the time dimension, which is helpful for multi-device data fusion and correlation analysis. After obtaining the standard sampling frequency, the original data stream is time-series aligned to unify the data from multiple devices to the same time step. Processing methods include interpolating sensors with uneven sampling intervals and filling missing data with the average or linear estimate of adjacent time points. The accuracy of time interpolation is typically controlled within ±1 millisecond to ensure that the dynamic characteristics of high-frequency vibration and current signals are not disrupted. For sensors with short-term frame drops or delays, continuity is maintained through smooth interpolation methods. After time alignment, the data forms a multidimensional matrix, with each row corresponding to a unified time point and each column representing the sensing channels of different devices. This step achieves time unification across devices, ensuring that vibration, temperature, current, and other sensor data are strictly aligned on the time axis, providing a reliable foundation for subsequent filtering, feature extraction, and fault analysis.

[0047] To ensure data quality, noise detection and filtering optimization are required for the time-series synchronous data stream. Noise primarily originates from sensor measurement errors, mechanical vibration interference, and communication acquisition jitter. Noise detection is achieved by analyzing the high-frequency components and signal-to-noise ratio of the signal; vibration signals with high-frequency jitter exceeding 10 Hz and amplitude exceeding 3 g are identified as abnormal noise. Filtering employs an adaptive Kalman filter combined with wavelet denoising. The Kalman filter dynamically adjusts the state estimation weights based on the variance of the sensor signal, while wavelet denoising smooths the signal against impulse interference and transient anomalies. After filtering, the high-frequency jitter of the vibration signal is reduced by approximately 60%, temperature signal fluctuations are reduced by approximately 50%, and the smoothness of the current signal is improved by over 30%. The filtered and optimized data stream retains dynamic characteristics while reducing noise interference, providing a high-quality data source for equipment condition monitoring and fault feature extraction. To analyze short-term changes in equipment condition, the filtered and denoised data stream is sliced ​​according to a set segmentation period. The segmentation period is set based on the equipment's operating cycle time; for example, it can be set to 10 to 30 seconds for continuous production line equipment, while a single processing cycle can be used as the time window for discrete processing equipment. During the segmentation process, a sliding window method is employed, with the window length consistent with the segmentation period and an overlap ratio typically between 25% and 40% to ensure the continuity of boundary data. Each time window data stream forms an independent segment, containing complete sampling data from each sensor channel, and allowing for the statistical analysis of mean, variance, peak value, and frequency domain characteristics within that time window. Multi-time window segmentation enables phased data processing and provides dynamic temporal characteristics for equipment fault trend analysis and prediction. Based on the multi-time window data stream, sensor data from different devices are centrally visualized. The visualization uses time as the horizontal axis and sensor channels or feature values ​​as the vertical axis, displaying dynamic data changes through trend curves, heatmaps, or energy distribution maps. For example, vibration signals are displayed using heatmaps to show amplitude changes, temperature signals using curves to show trends, and current changes using bar charts to show transient peak values. The visualization update frequency is controlled within 0.5 seconds to ensure real-time dynamic display. Through centralized visualization, the operating status, abnormal fluctuations, and potential fault points of each device can be intuitively presented, providing intuitive data support for industrial big data analysis, equipment fault prediction, and maintenance timing determination.

[0048] In this embodiment, see Figure 3 The diagram below illustrates the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include: The system performs operational status trend analysis on the visualized production data stream, extracting the changing trends of multiple operational indicators, including equipment vibration amplitude, speed fluctuation, current load, temperature change, pressure feedback, and power factor. Based on the visualized production data stream, the data operation baseline is calculated to obtain the equipment operation status baseline. The deviation of the change trend is evaluated based on the equipment's operating baseline. When the change trend reaches the third-level deviation standard, it is marked as a potentially abnormal device.

[0049] In this embodiment, after completing the centralized visualization of multi-device data, it is necessary to perform operational status trend analysis on the data stream to reveal the changing patterns of the equipment's dynamic performance over time. The analysis first processes key operational indicators, including vibration amplitude, speed fluctuation, current load, temperature change, pressure feedback, and power factor. Taking vibration amplitude as an example, by analyzing the mean, peak value, and spectral characteristics within different time windows, the vibration trends of the equipment in the low-frequency and high-frequency bands can be extracted. Typically, the low-frequency band reflects the overall state of the mechanical structure, while the high-frequency band reflects localized wear or loosening. Speed ​​fluctuation is captured by statistically analyzing periodic speed deviations, acceleration changes, and instantaneous speed gradients to identify changes in the equipment's stability under different load conditions. The current load indicator identifies excessive load or sudden fluctuations by calculating the difference between the mean and peak current values ​​for each time window. Temperature change analysis focuses on the hourly rise and fall of the spindle, bearings, and oil temperature ranges, with a typical allowable variation within ±2°C. Pressure feedback monitors the stability of the hydraulic and pneumatic systems; deviations exceeding 0.5 bar are considered abnormal. Power factor trend analysis reflects motor load matching and energy efficiency. By performing trend fitting and local slope analysis on the curve changes of each indicator over time, the real-time operating trend of each indicator can be obtained, providing basic data for deviation assessment. To effectively assess the equipment's operating status, a baseline for equipment operation needs to be established. The baseline is calculated using historical normal operating data as a reference, combined with the average value and variation amplitude of the current time window. For vibration amplitude, the moving average over the past 10 to 30 minutes is typically used as the baseline, allowing short-term fluctuations within ±0.5 g; the speed fluctuation baseline is centered on the equipment's rated speed, allowing a deviation of ±3% of the speed; the current load baseline references the equipment's rated power and load curve, allowing a deviation of ±5%; the temperature change baseline is set according to the equipment's normal operating range, for example, a bearing temperature within the range of 40 to 60°C is considered normal; the pressure feedback baseline is centered on the rated pressure of the hydraulic system, allowing a deviation of ±0.3 bar; the power factor baseline is determined based on the motor type and load level, typically within the range of 0.85 to 0.95. By comprehensively calculating multiple indicators, the baselines of each indicator are integrated into a reference curve for equipment operation, making the current operating status of the equipment comparable to the historical reference status, and providing a unified reference standard for subsequent deviation analysis.

[0050] After obtaining the baseline of equipment operation, deviation assessment of the trend changes of various operating indicators is required. Assessment methods include hourly deviation analysis, sliding window deviation accumulation, and multi-level threshold determination. First, the actual values ​​of each indicator in each time window are compared with the baseline to calculate the deviation amplitude and rate of change. For example, a vibration amplitude exceeding ±1 g of the baseline or exceeding ±0.8 g for three consecutive time windows is considered a Level 1 deviation; a speed fluctuation deviating from the baseline by more than ±5% is marked as a Level 2 deviation; a temperature fluctuation exceeding ±3°C or a pressure feedback exceeding ±0.5 bar is considered a Level 2 or Level 3 deviation. The assessment process simultaneously considers the duration and cumulative effect of deviations. Short-term fluctuations are not marked as abnormal, while indicators continuously exceeding thresholds trigger a potential anomaly warning mechanism. Multiple indicator deviations are quantified simultaneously to form a comprehensive deviation index, providing data for identifying abnormal equipment. Deviation analysis uses a combination of visualized trend curves and statistical tables to make the abnormal change patterns clear at a glance. When the deviation assessment result reaches the third-level deviation standard, i.e., both the deviation amplitude and duration exceed the set thresholds, the equipment is marked as potentially abnormal. The third-level deviation standard typically corresponds to high-risk areas for core equipment indicators, such as vibration amplitude continuously exceeding 2 g, speed fluctuation exceeding 10%, temperature exceeding the rated range by ±5°C, or pressure deviation exceeding ±0.7 bar. After marking potentially abnormal equipment, the type of abnormal indicator, deviation amplitude, deviation duration, and occurrence time can be recorded, providing data support for subsequent fault diagnosis and maintenance planning. Potential anomaly marking not only indicates potential faults in current equipment operation but also provides tagged data for industrial big data models, facilitating the prediction of future equipment failure trends and maintenance timing. The entire marking process is automated, using threshold triggering, trend accumulation, and multi-indicator fusion judgment to make equipment anomaly identification both scientific and reliable, providing high-value references for production safety management and maintenance scheduling.

[0051] In this embodiment, the specific steps for evaluating the deviation of the change trend based on the equipment operating baseline, and marking it as a potentially abnormal device when the change trend reaches the third-level deviation standard, are as follows: Three-layer deviation evaluation rules are set based on the equipment operating status baseline; The three-layer deviation evaluation rule is as follows: the first-layer deviation standard is: the deviation period is 10 minutes, and the parameter offset range is 10%; The second-level deviation standard is: a deviation period of 30 minutes and a parameter offset range of 20%. The third-level deviation standard is: a deviation period of 60 minutes and a parameter offset range of 30%. The change trend is evaluated for deviation according to the three-level deviation evaluation rule. When the change trend reaches the third-level deviation standard, it is marked as a potential abnormal device.

[0052] In this embodiment, after establishing the equipment operating baseline, it is necessary to formulate graded deviation assessment rules to quantify the degree of deviation between the equipment's operating state and the normal baseline. The deviation assessment rules are divided into three layers, each setting specific parameters for different degrees of operational anomalies and their durations. The first-layer deviation standard focuses on detecting short-term, minor deviations, setting a deviation period of 10 minutes and a parameter offset range of 10%, suitable for preliminary fluctuation monitoring of core indicators such as vibration amplitude, speed fluctuations, temperature changes, and current load. The second-layer deviation standard targets moderate-amplitude deviations, extending the deviation period to 30 minutes and setting the parameter offset range to 20%, which can identify performance degradation trends or periodic fluctuation anomalies in the medium-term operation of the equipment. The third-layer deviation standard is the highest level, with a deviation period of 60 minutes and a parameter offset range of 30%, used to capture long-term abnormal trends or potential faults with significant cumulative effects. Through the graded setting of these three layers of rules, both short-term fluctuations can be responded to quickly, and medium- and long-term trend anomalies can be identified, achieving multi-scale monitoring of the equipment's operating state. After the three-layer deviation rules are set, a layer-by-layer deviation analysis is performed on the changing trends of various operating indicators of the equipment. The analysis process begins with a first-level deviation assessment, statistically analyzing the mean, peak value, and magnitude of change for each indicator over a 10-minute period. If an indicator deviates from the baseline by more than ±10% within 10 minutes, it is recorded as a Level 1 deviation, and its trend is continuously monitored. Subsequently, data is accumulated to a second-level deviation assessment period of 30 minutes. The average deviation and maximum deviation magnitude of the indicator within this period are calculated using a sliding window of data over a continuous time period. If the indicator exceeds a ±20% offset range, it is marked as a Level 2 deviation. Finally, based on the third-level deviation rules, a long-term cumulative deviation analysis is performed on the trend over 60 minutes, focusing on whether the indicator exhibits sustained deviation or cumulative anomalies throughout the entire period. When the deviation magnitude exceeds ±30%, the highest-level deviation marker is triggered. The analysis process employs a sliding time window combined with a multi-indicator fusion method, ensuring that deviation assessment considers not only instantaneous amplitude but also the persistence and cumulative effect of the deviation, thereby accurately identifying potential abnormal trends in the equipment.

[0053] When equipment operating indicators reach the third-level deviation standard (parameter deviation exceeding ±30%) within a 60-minute deviation cycle, the equipment is marked as potentially abnormal. At this point, all core indicators, including vibration amplitude, speed fluctuation, current load, temperature change, pressure feedback, and power factor, need to be comprehensively analyzed to confirm the type, magnitude, and duration of the anomaly. After marking potentially abnormal equipment, the deviation start time, peak time, and deviation duration are recorded to provide data for subsequent fault location and maintenance time prediction. For example, if the bearing vibration amplitude deviates from the baseline by more than 20 g on average within 60 minutes, and the temperature rises by more than 3°C, the bearing is considered to have a potential anomaly risk, triggering an early warning. Potential anomaly marking is not only used for real-time monitoring but also provides tagged data for industrial big data analysis, which can be used to train equipment fault prediction models or optimize maintenance plans. Through hierarchical deviation rules and cumulative deviation analysis, short-term fluctuations, periodic anomalies, and potential faults can be effectively distinguished, providing scientific decision support for factory equipment maintenance.

[0054] In this embodiment, step S3 includes the following steps: Continuously track the status changes of potentially abnormal equipment and record the deviation value tracking sequence; The rate and magnitude of change are calculated based on the deviation value tracking sequence. Based on the rate and magnitude of change, monotonic growth is confirmed, and deviation trend analysis is performed to generate equipment deviation status. Based on the equipment deviation situation, identify the deviation growth trend, locate the faulty equipment, and mark the faulty equipment.

[0055] In this embodiment, after potentially abnormal equipment is marked, its operating status needs to be continuously monitored to form a deviation value tracking sequence. The tracking process focuses on key indicators of the equipment, including vibration amplitude, speed fluctuation, current load, temperature change, pressure feedback, and power factor. During the tracking period, the data for each indicator is recorded in a time series format, and the deviation value is calculated as the difference between the actual value and the equipment's operating baseline. The recording time step is consistent with the previous standard sampling frequency, typically 50 to 100 milliseconds, to ensure the continuity of highly dynamic indicators (such as vibration and current). During the tracking process, each data point includes a timestamp, deviation amplitude, and indicator type, and is stored as a continuous sequence for subsequent rate of change and trend analysis. This method can completely capture the evolution of the equipment's state from the time of potential anomaly marking to the current time, providing refined and dynamic input data for deviation trend analysis. After obtaining the deviation value tracking sequence, it is necessary to further quantify the dynamic characteristics of the equipment's state changes, including the rate of change and the magnitude of change. The rate of change is obtained by calculating the time difference of the deviation values ​​at consecutive sampling points, reflecting how quickly the equipment's condition changes. For example, a vibration amplitude change rate of 1 to 2 g / s is considered a moderate rate, while a rate exceeding 3 g / s is considered a rapid change. The amplitude of change is calculated by summing the maximum and minimum deviation values ​​within each tracking cycle, used to measure the absolute extent to which the equipment indicators deviate from the baseline. For example, a maximum deviation of +5°C in bearing temperature and a maximum deviation of 0.6 bar in pressure feedback within 60 minutes indicate potential cumulative anomalies in the equipment. By calculating both the rate of change and the amplitude, the dynamic change characteristics of each indicator can be established, facilitating further trend determination and potential fault analysis.

[0056] Using data on rate of change and amplitude, monotonically increasing deviation sequences of potentially abnormal equipment are confirmed and trend analyzed. Monotonically increasing confirmation involves determining whether the deviation sequence exhibits a continuous upward or downward trend, excluding the influence of short-term fluctuations and random disturbances. For example, if the vibration amplitude maintains a monotonically increasing trend for 30 consecutive minutes and the cumulative amplitude exceeds 1.5 g, the deviation can be considered to be showing a monotonically increasing trend. Trend analysis employs a sliding time window method, with a window length of 10 to 15 minutes, calculating the average growth rate and standard deviation of the deviation within the window to determine whether the deviation change is significant. The trends of various indicators are combined to generate an equipment deviation status, including the degree of deviation accumulation, growth rate, and potential anomaly level. For example, a rapid increase in vibration amplitude, a slow increase in temperature, and increased current load fluctuations generate a comprehensive deviation status label of "high-risk cumulative growth." This equipment deviation status provides a dynamic and quantifiable basis for further location of abnormal equipment and fault identification. After obtaining the equipment deviation status, the overall trend of deviation growth needs to be identified to determine the location and specific indicators of the faulty equipment. Deviation growth status identification is completed through comprehensive analysis of the cumulative deviation and rate of change of multiple indicators. When equipment exhibits a sustained monotonically increasing deviation of at least two core indicators (such as vibration amplitude and temperature) with a cumulative deviation exceeding 30% over a continuous 60-minute period, it is determined to be in the early stage of failure. Fault location is achieved by analyzing the spatial distribution of indicators and their correlation with the equipment structure. For example, if both bearing vibration and bearing temperature are abnormal, the fault can be located at the specific bearing unit; if the motor power factor decreases and is accompanied by abnormal current load, the fault can be located at the motor drive section. Faulty equipment is marked in the data records, including the specific equipment number, abnormal indicators, cumulative deviation value, and occurrence time. After marking, maintenance reminders or repair plans can be triggered, while also providing tagged fault data for industrial big data analysis, used for subsequent fault prediction model training and repair time prediction. Through continuous tracking, trend analysis, and multi-indicator fusion, accurate location of potentially abnormal equipment and early fault warnings can be achieved.

[0057] In this embodiment, step S4 includes the following steps: Perform production chain analysis on faulty equipment and extract the production chain; Identify upstream and downstream equipment based on the production chain and mark related upstream and downstream equipment; Based on upstream and downstream related equipment, multi-dimensional analysis of deviation parameters of faulty equipment is performed to generate multi-parameter linkage deviation data. Based on multi-parameter linkage deviation data, we perform equipment deviation attribution analysis to identify equipment failure factors.

[0058] In this embodiment, after identifying the faulty equipment, the first step is to perform production chain analysis to clarify the specific stage and related processes in the production process. Production chain analysis is achieved by acquiring information on the equipment's operational sequence, transmission technology, and material flow on the production line. Specific methods include comprehensive analysis of equipment sensor data, operation logs, and visual recognition data, mapping material flow, process sequence, and equipment operating status onto a production chain diagram. During production chain extraction, key parameters such as equipment operating cycle, operating time window, output quantity, and material transmission time are considered. Through time series and process node analysis of these parameters, a complete chain from upstream raw material processing to downstream finished product output can be drawn, and the workload and key performance indicators of each piece of equipment can be marked. This approach not only clarifies the faulty equipment's position in the overall production chain but also provides foundational data for upstream and downstream equipment correlation analysis. Refined extraction of the production chain supports subsequent multi-parameter deviation analysis and fault attribution, enabling the ability to trace the root cause of the fault. After establishing the production chain, it is necessary to identify upstream and downstream equipment that are directly or indirectly related to the faulty equipment. Upstream equipment refers to key processing nodes that precede the faulty equipment in the material or information flow, while downstream equipment refers to equipment nodes that receive the output of the faulty equipment and depend on its operational results. Upstream and downstream identification employs a combination of topological relationship analysis and time-series correlation analysis. First, the flow of materials and information is clarified through topological relationships, mapping production link nodes to equipment numbers and marking the preceding and following equipment for each node. Second, time-series cross-validation is performed using sensor data and work logs to calculate the correlation of indicators, such as vibration amplitude, temperature fluctuations, or the delay relationship of production cycles, to confirm the strength of upstream and downstream influences. After marking upstream and downstream equipment, a local production network diagram centered on the faulty equipment is formed, providing the target equipment range and data acquisition boundaries for multidimensional deviation analysis.

[0059] After identifying upstream and downstream related equipment, multidimensional analysis of the deviation parameters of the faulty equipment and its related equipment is required to generate multi-parameter linkage deviation data. Multidimensional analysis considers the interaction effects of key operating indicators, such as the changes in vibration amplitude, current load, speed fluctuation, temperature change, pressure feedback, and power factor among different equipment. Analysis methods include time-series correlation analysis, lag correlation calculation, and multivariate covariance analysis. For example, when the upstream motor load fluctuates significantly, the vibration of the downstream bearing may increase accordingly. Time lag window analysis can determine the impact delay, typically set to 30 to 60 seconds. Multidimensional analysis generates a multi-parameter linkage deviation data matrix, where each row represents a time node, and each column contains the deviation amplitude and rate of change of the indicators for the faulty equipment and its upstream and downstream equipment, thus quantifying the linkage deviation relationship between equipment and providing a basis for deviation attribution. After obtaining the multi-parameter linkage deviation data, equipment deviation attribution analysis is required to clarify the causes of the fault and influencing factors. Attribution analysis methods include causal relationship analysis, correlation matrix calculation, and deviation contribution rate assessment. First, time-series causal analysis is used to identify the main driving equipment for the deviation. For example, if the upstream motor current load is consistently high while the vibration amplitude of the faulty equipment cumulatively increases, the motor load fluctuation can be identified as the main factor causing the vibration anomaly. Second, by calculating the contribution rate of each indicator to the deviation change, the influence of different equipment or parameters on the occurrence of the fault is quantified. For example, vibration amplitude accounts for up to 60% of the bearing life, while temperature change accounts for approximately 20%. Finally, the analysis results are compiled into a list of equipment failure factors, including direct failure factors, indirect influencing factors, and their deviation amplitude, rate of change, and duration. This method can accurately locate the root cause of the fault from multiple equipment and parameter dimensions, providing a scientific basis for equipment maintenance planning and preventive maintenance, and providing high-quality feature input for equipment failure prediction models based on industrial big data.

[0060] In this embodiment, the specific steps of S5 are as follows: Based on the equipment deviation trend, a multi-indicator assessment of equipment health status is performed to generate an equipment health status curve. Based on equipment failure factors, the equipment health status curve is simulated to generate a failure evolution diagram. Predict equipment failure downtime based on equipment failure evolution diagram and output downtime time points; Based on the downtime point, the minimum maintenance cost is calculated and the equipment maintenance priority is planned to generate the equipment maintenance sequence.

[0061] In this embodiment, after obtaining the equipment deviation status, a multi-indicator comprehensive evaluation of the equipment health status is required to generate an equipment health status curve. Evaluation indicators include vibration amplitude, speed fluctuation, current load, temperature change, pressure feedback, and power factor, covering key equipment performance and operational stability. The health status evaluation employs indicator normalization and weighted methods to convert the deviation values ​​of each indicator into health scores. The score range is typically set from 0 to 100 points, where 100 points indicates the equipment is fully healthy, and 0 points indicates the equipment is about to fail. For example, a vibration amplitude deviation within 10% of the baseline scores 90 points, a 20% deviation scores 70 points, and a 30% deviation scores 50 points; a temperature change controlled within ±2°C scores 95 points, and a deviation exceeding ±5°C drops to 50 points. The weighted average of each indicator generates a curve showing the comprehensive health score changing over time, i.e., the equipment health status curve. This curve visually reflects the operational health level of the equipment at different time periods, providing a quantitative basis for subsequent fault evolution analysis and prediction. By utilizing previously identified equipment failure factors, a failure development simulation is performed on the health status curve to generate an equipment failure evolution diagram. The simulation method combines the dynamic trends of key equipment indicators, the rate of deviation growth, and historical failure modes to predict the evolution from a healthy state to a failure state. For example, if the bearing vibration amplitude increases continuously at a rate of 0.5 g per minute, while the temperature rises slowly, and the cumulative deviation reaches 3°C per hour, the simulation algorithm predicts that the bearing will reach the failure threshold within the next 12 hours. The failure development simulation employs a multi-factor cumulative model, superimposing the cumulative effects of deviations from various key indicators, while also considering the mutual influence between indicators; for example, abnormal vibration amplitude may accelerate temperature rise or current load fluctuations. The generated equipment failure evolution diagram uses time as the horizontal axis and the health status score as the vertical axis, marking potential future failure time windows and the evolution trends of key indicators, enabling maintenance personnel to intuitively understand the entire process of equipment evolution from the current state to a failure state.

[0062] After obtaining the equipment failure evolution diagram, equipment failure downtime is predicted using threshold determination and trend extrapolation methods. Downtime prediction is primarily based on the time point when the health status curve declines to a preset critical threshold. For example, a drop in the overall health score to 50 points or a core indicator deviation exceeding 30% is considered a critical downtime point. The prediction method incorporates the rate of indicator change, the cumulative trend of deviation, and historical data on similar equipment failures for correction, thereby improving prediction accuracy. For highly dynamic equipment, such as spindles or large motors, a short-term extrapolation window (30 minutes to 1 hour) can be used to predict future deviation trends, combined with long-term evolution trends (6 to 12 hours) for comprehensive downtime prediction. The output downtime points not only indicate the specific time but also provide key driving indicators and their deviation amplitudes, providing accurate basis for maintenance planning and effectively reducing the production risk of sudden downtime. Based on the predicted downtime points, an optimal maintenance strategy needs to be developed to achieve a balance between minimum maintenance costs and production impact. Minimum maintenance cost calculation considers factors such as maintenance manpower, materials, downtime losses, and spare parts usage costs. For example, bearing replacement costs 500 yuan, and maintenance downtime losses are approximately 2000 yuan / hour based on the equipment's hourly output value. Equipment maintenance priority planning is performed based on predicted downtime, equipment importance, and maintenance costs. Prioritization planning uses a sorting algorithm to prioritize equipment with high failure risk, imminent downtime, and low maintenance costs, while also considering production line continuity to avoid disruptions in critical links. The final generated equipment maintenance sequence includes equipment number, predicted downtime, maintenance priority, and estimated maintenance duration. For example, if the main motor's predicted downtime is within 12 hours, the maintenance duration is 2 hours, and the cost is moderate, it is assigned as the first priority; if the downstream conveyor equipment's predicted downtime is 24 hours, and the maintenance cost is low, it is assigned as the second priority. This method enables scientific scheduling of equipment maintenance, reduces downtime losses, and provides quantitative data for predictive maintenance driven by industrial big data.

[0063] In this embodiment, the specific steps for calculating the minimum maintenance cost based on the downtime point and planning the equipment maintenance priority to generate the equipment maintenance sequence are as follows: Extract the factory production schedule; perform task time-series sorting analysis based on the factory production schedule to extract the task sequence; Based on the downtime point and task sequence, the sudden downtime loss cost, direct maintenance cost, and associated equipment cost of downtime maintenance are calculated to obtain a three-level cost function; The minimum maintenance cost is calculated based on the three-layer cost function to determine the minimum maintenance time point; Based on the minimum maintenance time point, equipment maintenance priority is planned to generate an equipment maintenance sequence.

[0064] In this embodiment, before optimizing equipment maintenance, a complete factory production schedule needs to be obtained to clarify equipment operating times, process sequences, and capacity arrangements. The production schedule includes key information such as daily or weekly production plans, equipment operating hours, processing sequence, material requirements, and output quantities. For example, in the main production line schedule, the spindle processing equipment operates for 8 hours daily, while secondary processing equipment and auxiliary conveying equipment are scheduled to operate at 10-minute intervals. Schedule extraction obtains production information through a data interface and verifies the schedule data by combining it with a visual recognition system to monitor the actual operation of the equipment, ensuring the accuracy of the operation sequence and time nodes. The schedule information not only records equipment operating times but also includes the processing load and task priority of the equipment in each process, providing basic data for subsequent task timing analysis and cost calculation. After obtaining the schedule data, task timing sorting analysis is required to generate equipment task sequences. The task sequences reflect the sequential operation relationships and time dependencies of equipment in the production process. The sorting analysis combines the start and end times of operations, process dependencies, and equipment load in the schedule to arrange tasks for the same equipment and related equipment according to time sequence and process order. For example, for conveying equipment, the task sequence needs to consider the upstream raw material processing time and the downstream finished product processing time; for critical processing equipment, the task sequence needs to ensure that the main process is completed first to avoid production bottlenecks. The sorting analysis results generate an equipment task time sequence table, including task number, start and end time, task duration, and task priority, providing a clear time reference for calculating downtime costs and determining the minimum maintenance time.

[0065] Based on the task sequence, it is necessary to quantify the economic impact of equipment maintenance and downtime by establishing a three-layer cost function. The first layer represents the cost of unexpected downtime, calculating the production loss and delayed delivery losses caused by equipment downtime not being planned. For example, the downtime loss for a spindle is approximately 2000 yuan per hour. The second layer represents the direct costs of maintenance, including labor, materials, and spare parts costs. For example, bearing replacement costs 500 yuan, and maintenance labor costs 300 yuan. The third layer represents the costs of related equipment affected by downtime maintenance, considering the capacity loss of upstream and downstream equipment due to downtime. For example, if the spindle is down for 2 hours, the downstream conveyor equipment loses approximately 1000 yuan in output value. By summing these three layers of costs to form the three-layer cost function, the economic impact of different downtime points and different equipment maintenance plans can be quantified, providing data for minimum cost optimization. Using the aforementioned three-layer cost function, minimum maintenance cost calculation can be performed, thereby determining the optimal maintenance time. The calculation method analyzes the changes in total cost at different downtime points and selects the time with the minimum total cost as the optimal maintenance time. For example, if the spindle equipment is repaired 6 hours before the predicted downtime, the total cost is approximately 3,500 yuan; if repaired immediately after downtime, the total cost is approximately 4,800 yuan. Considering the production task sequence and equipment load, the time point with the lowest cost, 3 hours in advance, is selected for repair, achieving a balance between downtime losses and repair costs. The calculation process can also consider equipment importance coefficients and production continuity constraints to avoid delays in critical processes due to early repairs, ensuring that the minimum repair cost calculation is both scientific and feasible. After determining the optimal repair time, an equipment repair priority plan needs to be developed to generate a complete equipment repair sequence. Priority planning considers equipment failure risk, downtime proximity, repair costs, and the criticality of production tasks. High-risk equipment with imminent downtime and lower repair costs is prioritized, while low-risk equipment or equipment with distant downtime is prioritized. For example, if the spindle equipment's downtime is expected to be within 6 hours and the core indicator deviation exceeds 30%, its priority is set to first; if the downstream conveyor equipment's downtime is 24 hours and the deviation is smaller, its priority is set to third. The final generated equipment maintenance sequence includes equipment number, optimal maintenance time, maintenance priority, estimated maintenance duration and related cost information, providing an executable solution for production scheduling and maintenance arrangements, and achieving a balance between production continuity and maintenance economy.

[0066] In this embodiment, a prediction system based on industrial big data is provided for executing the prediction method based on industrial big data as described above, including: The cycle segmentation module is used to extract equipment data streams from the factory production process, perform multi-cycle segmentation and centralized visualization, and generate a visualized production data stream. The deviation assessment module is used to assess deviations based on the visualized production data stream and mark potentially abnormal equipment. The fault location module is used to calculate the rate and magnitude of change of the potentially abnormal equipment; locate the faulty equipment based on the rate and magnitude of change; and mark the faulty equipment. The attribution analysis module is used to perform multi-parameter linkage analysis and deviation attribution analysis on faulty equipment to identify the factors causing equipment failure. The downtime prediction module is used to predict equipment failure downtime and plan equipment maintenance priorities based on equipment failure factors, and generate equipment maintenance sequences.

[0067] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0068] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein are implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A prediction method based on industrial big data, characterized in that, Includes the following steps: Step S1: Extract the equipment data stream of the factory production process, divide it into multiple cycles and centrally visualize it to generate a visualized production data stream; Step S2: Perform deviation assessment based on the visualized production data stream and mark potentially abnormal equipment; Step S3: Calculate the rate and magnitude of change of the potentially abnormal equipment; locate the faulty equipment based on the rate and magnitude of change, and mark the faulty equipment; Step S4: Perform multi-parameter linkage analysis and deviation attribution analysis on the faulty equipment to identify the factors causing the equipment failure; Step S5: Based on equipment failure factors, predict equipment failure downtime and plan equipment maintenance priorities to generate an equipment maintenance sequence.

2. The prediction method based on industrial big data according to claim 1, characterized in that, The specific steps of step S1 are as follows: Extracting equipment data streams from the factory production process based on multiple device sensors; Determine the data sampling frequency of the multi-device sensors; calculate the average frequency based on the data sampling frequency to generate a standard sampling frequency; The device data stream is time-aligned according to the standard sampling frequency to generate a time-synchronized data stream. Sensor noise detection and adaptive filtering optimization are performed on the time-synchronized data stream to generate a filtered and noise-reduced data stream. Define a segmentation period, and divide the filtered and denoised data stream into multiple periods based on the segmentation period to generate multiple time window data streams; Centralized visualization of multi-device data streams across multiple time windows generates a visualized production data stream.

3. The prediction method based on industrial big data according to claim 1, characterized in that, The specific steps of step S2 are as follows: The system performs operational status trend analysis on the visualized production data stream, extracting the changing trends of multiple operational indicators, including equipment vibration amplitude, speed fluctuation, current load, temperature change, pressure feedback, and power factor. Based on the visualized production data stream, the data operation baseline is calculated to obtain the equipment operation status baseline. The deviation of the change trend is evaluated based on the equipment's operating baseline. When the change trend reaches the third-level deviation standard, it is marked as a potentially abnormal device.

4. The prediction method based on industrial big data according to claim 3, characterized in that, The specific steps for evaluating the deviation of the change trend based on the equipment operating status baseline, and marking the change trend as a potentially abnormal device when it reaches the third-level deviation standard, are as follows: Three-layer deviation evaluation rules are set based on the equipment operating status baseline; The three-layer deviation evaluation rule is as follows: the first-layer deviation standard is: the deviation period is 10 minutes, and the parameter offset range is 10%; The second-level deviation standard is: a deviation period of 30 minutes and a parameter offset range of 20%. The third-level deviation standard is: a deviation period of 60 minutes and a parameter offset range of 30%. The change trend is evaluated for deviation according to the three-level deviation evaluation rule. When the change trend reaches the third-level deviation standard, it is marked as a potential abnormal device.

5. The prediction method based on industrial big data according to claim 1, characterized in that, Step S3 is as follows: Continuously track the status changes of potentially abnormal equipment and record the deviation value tracking sequence; The rate and magnitude of change are calculated based on the deviation value tracking sequence. Based on the rate and magnitude of change, monotonic growth is confirmed, and deviation trend analysis is performed to generate equipment deviation status. Based on the equipment deviation situation, identify the deviation growth trend, locate the faulty equipment, and mark the faulty equipment.

6. The prediction method based on industrial big data according to claim 1, characterized in that, The specific steps of step S4 are as follows: Perform production chain analysis on faulty equipment and extract the production chain; Identify upstream and downstream equipment based on the production chain and mark related upstream and downstream equipment; Based on upstream and downstream related equipment, multi-dimensional analysis of deviation parameters of faulty equipment is performed to generate multi-parameter linkage deviation data. Based on multi-parameter linkage deviation data, we perform equipment deviation attribution analysis to identify equipment failure factors.

7. The prediction method based on industrial big data according to claim 1, characterized in that, The specific steps of step S5 are as follows: Based on the equipment deviation trend, a multi-indicator assessment of equipment health status is performed to generate an equipment health status curve. Based on equipment failure factors, the equipment health status curve is simulated to generate a failure evolution diagram. Predict equipment failure downtime based on equipment failure evolution diagram and output downtime time points; Based on the downtime point, the minimum maintenance cost is calculated and the equipment maintenance priority is planned to generate the equipment maintenance sequence.

8. The prediction method based on industrial big data according to claim 7, characterized in that, The specific steps for calculating the minimum maintenance cost based on downtime points and planning equipment maintenance priorities to generate an equipment maintenance sequence are as follows: Extract the factory production schedule; Based on the factory production schedule, task time sequence sorting analysis is performed to extract task sequences; Based on the downtime point and task sequence, the sudden downtime loss cost, direct maintenance cost, and associated equipment cost of downtime maintenance are calculated to obtain a three-level cost function; The minimum maintenance cost is calculated based on the three-layer cost function to determine the minimum maintenance time point; Based on the minimum maintenance time point, equipment maintenance priority is planned to generate an equipment maintenance sequence.

9. A prediction system based on industrial big data, characterized in that, The method for performing the prediction method based on industrial big data as described in claim 1 includes: The cycle segmentation module is used to extract equipment data streams from the factory production process, perform multi-cycle segmentation and centralized visualization, and generate a visualized production data stream. The deviation assessment module is used to assess deviations based on the visualized production data stream and mark potentially abnormal equipment. The fault location module is used to calculate the rate and magnitude of change of the potentially abnormal equipment; locate the faulty equipment based on the rate and magnitude of change; and mark the faulty equipment. The attribution analysis module is used to perform multi-parameter linkage analysis and deviation attribution analysis on faulty equipment to identify the factors causing equipment failure. The downtime prediction module is used to predict equipment failure downtime and plan equipment maintenance priorities based on equipment failure factors, and generate equipment maintenance sequences.

Images