Method for early warning of abnormality of continuous casting equipment based on current change trend

Abstract

The application discloses a method for early warning of abnormality of continuous casting equipment based on current change trend, which collects equipment fault information, collects current data of the equipment with faults and the equipment without faults as selected training set, analyzes the change law of the current data, and proposes a trend alarm mechanism; if the current value is increased in a short period, the trend alarm mechanism captures the current cumulative increase value in the short period; if the current value is slowly and steadily increased in a long period, the trend alarm mechanism captures the number of continuous days of increase and the cumulative increase amount of the current in the period; the over-limit condition characteristic value of the trend alarm mechanism is set, the current data of the equipment is collected and calculated, the over-limit condition characteristic value is met, and equipment abnormal early warning is given. The method monitors the current change trend of the equipment, constructs an early warning model based on the current change trend, compares and analyzes the data of multiple equipment, improves the accuracy and universality of the early warning model, and accurately early warns the equipment fault hidden danger.

Description

Technical Field

[0001] This invention relates to the field of equipment monitoring and diagnostic technology, and in particular to a method for early warning of abnormalities in continuous casting equipment based on current change trends. Background Technology

[0002] As enterprise informatization deepens, a large number of information systems are widely used in enterprises. Technologies such as the Internet of Things, cloud computing, and industrial internet are closely integrated with enterprise operations and production. Data collection processes such as equipment operation, production and processing, and testing are becoming more automated. Enterprises have accumulated a large amount of data, and all stages of enterprise operations and production can be recorded. These are all important assets of enterprises.

[0003] With the continuous development of enterprise operations and production, trends have become a very important concept and a focus of attention. Trend analysis compares the same indicators or ratios in data from different periods, directly observing their increases or decreases and the magnitude of these changes to examine their development trends and predict their future prospects. It is based on the principle of the continuity of things over time to predict their development trends. Trend analysis can provide statistical support for a comprehensive analysis of targets and can provide early warnings of potential risks. Trend analysis methods can be broadly classified into longitudinal analysis, cross-sectional analysis, standard analysis, and comprehensive analysis. Trend prediction typically uses regression analysis, exponential smoothing, and other methods to analyze and predict relevant indicator data, analyze their development trends, and predict possible outcomes.

[0004] For equipment monitoring and diagnosis, current methods often rely on single-parameter over-limit warnings to determine equipment faults. However, due to factors such as operating condition interference, data fluctuations are significant, leading to numerous false alarms when using single-parameter over-limit warnings and failing to detect equipment faults in a timely manner. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a method for early warning of abnormalities in continuous casting equipment based on current change trends. This method overcomes the shortcomings of traditional single-parameter over-limit early warning for judging equipment failures. By monitoring the current change trend of the equipment, an early warning model based on the current change trend is constructed. By comparing and analyzing data from multiple equipment, the accuracy and versatility of the early warning model are improved, accurately warning of potential equipment failures, improving the reliability of equipment operation, and ensuring the smooth operation of production.

[0006] To solve the above-mentioned technical problems, the present invention provides a method for early warning of abnormalities in continuous casting equipment based on current change trends, comprising the following steps:

[0007] Step 1: Collect equipment fault information and collect current data from faulty and fault-free equipment as the selected training set;

[0008] Step 2: Analyze the changing patterns of current data, filter out current values ​​in the shutdown state, and propose the mechanism for trend alarm;

[0009] Step 3: If the current change trend is characterized by an increase in current value within a short period, the trend alarm mechanism will capture the current by controlling the cumulative increase in current within a short period. If the current change trend is characterized by a slow and stable increase in current value within a long period, the trend alarm mechanism will capture the current by controlling the number of consecutive days of increase and the cumulative increase in current during this period.

[0010] Step 4: Set the over-limit condition characteristic value of the trend alarm mechanism: the difference of the daily average equipment current is >0 and the difference of the daily average equipment current within the consecutive m days is >a, where m is the number of consecutive days, a is the threshold of the difference of the daily average equipment current, and m and a are adjustable parameters.

[0011] Step 5: Collect and calculate equipment current data, identify over-limit condition characteristic values, and provide an early warning of equipment anomalies.

[0012] Furthermore, the average current data of the equipment is calculated daily. The trend of current change is judged based on the change of the average value between two consecutive days. A difference of more than 0 between the daily average values ​​indicates an increase, and a difference of less than 0 indicates a decrease.

[0013] Furthermore, for devices where the current value increases within a short period, the value of m ranges from 2 to 3 with a step size of 1, and the value of a ranges from 1 to 9 with a step size of 1; for devices where the current value increases slowly and steadily within a long period, the value of m ranges from 3 to 8 with a step size of 1, and the value of a ranges from 0.1 to 1.9 with a step size of 0.1.

[0014] Furthermore, for devices with rising current values ​​within a short period, m=3 and a=5 are set, meaning the current continues to rise for 3 days, with a cumulative increase of 5 during this period; for devices with slowly and steadily rising current values ​​within a long period, m=8 and a=1.3 are set, meaning the current continues to rise for 8 days, with a cumulative increase of 1.3 during this period.

[0015] This invention employs the aforementioned technical solution for early warning of continuous casting equipment anomalies based on current change trends. Specifically, this method collects equipment fault information, using current data from both faulty and fault-free equipment as a training set; analyzes the changing patterns of the current data, and proposes a trend-based alarm mechanism. If the current value increases within a short period, the trend-based alarm mechanism captures this increase by controlling the cumulative current rise within that short period; if the current value rises slowly and steadily over a long period, the trend-based alarm mechanism captures this increase by controlling the number of consecutive days of increase and the cumulative current rise during that period. It sets over-limit condition characteristic values ​​for the trend-based alarm mechanism, collects and calculates equipment current data, and provides an early warning of equipment anomalies when the over-limit condition characteristic values ​​are met. This method overcomes the shortcomings of traditional single-parameter over-limit early warning systems that rely on specific equipment fault identification. By monitoring the current change trend of the equipment, it constructs an early warning model based on the current change trend. Through comparative analysis of data from multiple equipment units, it improves the accuracy and versatility of the early warning model, accurately predicting potential equipment faults, improving the reliability of equipment operation, and ensuring smooth production. Detailed Implementation

[0016] The method for early warning of continuous casting equipment malfunctions based on current change trends, as described in this invention, includes the following steps:

[0017] Step 1: Collect equipment fault information and collect current data from faulty and fault-free equipment as the selected training set;

[0018] Step 2: Analyze the changing patterns of current data, filter out current values ​​in the shutdown state, and propose the mechanism for trend alarm;

[0019] Step 3: If the current change trend is characterized by an increase in current value within a short period, the trend alarm mechanism will capture the current by controlling the cumulative increase in current within a short period. If the current change trend is characterized by a slow and stable increase in current value within a long period, the trend alarm mechanism will capture the current by controlling the number of consecutive days of increase and the cumulative increase in current during this period.

[0020] Step 4: Set the over-limit condition characteristic value of the trend alarm mechanism: the difference of the daily average equipment current is >0 and the difference of the daily average equipment current within the consecutive m days is >a, where m is the number of consecutive days, a is the threshold of the difference of the daily average equipment current, and m and a are adjustable parameters.

[0021] Step 5: Collect and calculate equipment current data, identify over-limit condition characteristic values, and provide an early warning of equipment anomalies.

[0022] Preferably, the average current data of the device is calculated daily, and the trend of current change is judged to be rising or falling based on the change of the average value of two adjacent days. A difference of more than 0 between the daily average values ​​indicates an increase, and a difference of less than 0 indicates a decrease.

[0023] Preferably, for devices where the current value increases within a short period, the value of m ranges from 2 to 3 with a step size of 1, and the value of a ranges from 1 to 9 with a step size of 1; for devices where the current value increases slowly and steadily within a long period, the value of m ranges from 3 to 8 with a step size of 1, and the value of a ranges from 0.1 to 1.9 with a step size of 0.1.

[0024] Preferably, for devices with a short-term current increase, m=3 and a=5 are set, meaning the current continues to rise for 3 days and the cumulative increase during this period is 5; for devices with a long-term current increase that is slow and stable, m=8 and a=1.3 are set, meaning the current continues to rise for 8 days and the cumulative increase during this period is 1.3.

[0025] When equipment deteriorates, although current data may fluctuate due to various reasons, the overall trend is upward. This method uses longitudinal trend analysis to analyze and warn of this trend, which can effectively avoid the interference of current data fluctuations and promptly detect equipment failures. Due to the differences between equipment, a trend warning model set for a single device may not be universally applicable to other devices. This method utilizes data from a large number of devices to uncover hidden value and trains a universal trend warning model through extensive data. This model can then be quickly applied to a large number of devices, ensuring model accuracy while saving the time and manpower costs of adjusting models one by one.

[0026] This method applies an early warning model based on current change trends to monitor the current change trends of equipment. By comparing and analyzing data from multiple devices, the accuracy and versatility of the model are improved, and timely reminders are given to troubleshoot potential equipment malfunctions on-site. For example, it can address issues such as broken rolls in the continuous casting sector or poor motor lubrication, thereby improving the reliability of equipment operation and ensuring smooth production.

[0027] This method categorizes trend warning mechanisms into two types: short-term sharp increases and long-term slow increases. Different trend warning mechanisms are proposed and models are built for each type, and large datasets are used to train the models and select appropriate pre-emptive parameter settings. While the models for both types of warning mechanisms are essentially the same, individual differences exist between devices, manifested in the presence of some non-fixed parameter characteristics in the models. Adjusting these parameter characteristics to adapt to the specific application needs of different devices is a conventional "one model per device" approach. To achieve greater model versatility, the most suitable parameter characteristic settings need to be found through model training; this approach belongs to supervised learning. This method first selects a batch of faulty and fault-free devices as a training set, determining the most suitable parameter characteristic values ​​based on real data to effectively distinguish between faulty and fault-free devices. Then, a different batch of data is selected as a validation set, and the parameter characteristics are recalculated and compared with the values ​​determined during training. If the model can still effectively distinguish between faulty and fault-free devices, the versatility of the model's parameter characteristics is verified.

[0028] In a practical application of this method, two equipment failures occurred in the sector section of a steel plant, as shown in Table 1. The failures were a broken roller and poor lubrication. The current data before the failures both showed an upward trend, which can be used to alert the site to troubleshoot potential equipment failures.

[0029] Table 1

[0030] Equipment Name Fault type Failure time 14-segment motor Poor lubrication leads to high motor load and high current. 2022 / 6 / 17 7-segment lower motor Broken Roller 2022 / 6 / 12

[0031] We selected 121 motors in the sector segment and extracted 121 current data points from June to July 2022, including two faulty devices and 119 fault-free devices.

[0032] Analyze the current trends of the two faulty devices and describe them using different mathematical methods;

[0033] For motors with 14 segments, the trend is characterized by a high increase in current value in the short term. For this type of data, the upward trend is mainly captured by controlling the cumulative increase value in the short term.

[0034] For a 7-segment motor, the trend is characterized by a slow and stable increase in current value over a long period. For this type of data, the upward trend is mainly controlled by the number of consecutive days of increase and the cumulative increase in current during this period.

[0035] To describe the trend change, two characteristic values ​​are selected as alarm criteria. The first criterion is the number of times the daily average difference is consecutively greater than 0. To ensure a significant increase, the increase in the daily average is also set as the second criterion. By controlling different values ​​for these two criteria, the goal of providing early warning of an upward trend is achieved. This ensures that the warning can detect faulty equipment while preventing missed or false alarms.

[0036] Comparing the results of different warning conditions, with the goal of capturing two faulty devices separately, the optimal warning condition selected on the training set is:

[0037] (1) Trend 1: The current value rises significantly in the short term, with a continuous rise of 3 days and a cumulative increase of 5. Under this setting, accurate warnings can be issued on all 121 training data points without any missed or false alarms.

[0038] (2) Trend 2: The current value rises slowly and steadily over a long period of time, with a continuous rise of 8 days and a cumulative increase of 1.3. Under this setting, accurate warnings were issued on all 121 training data points without any missed or false alarms.

[0039] Methods for training models with data:

[0040] 1) Trend 1: The current value rises significantly in the short term. Based on two months of current data from 121 devices in the training set, two characteristic values, m and a, were calculated for each device. Adjustment methods for m and a were selected based on their range and magnitude of change. The value of m ranged from 2 to 3 with a step size of 1, and the value of a ranged from 1 to 9 with a step size of 1. Each adjustment of m and a formed a combination, which was used to trigger an alarm. The number of devices triggering alarms under different combinations was counted, as shown in Table 2.

[0041] Table 2

[0042]

[0043] Table 2 lists the values ​​of one parameter selected from each row and column. The combined result is the number in the table. For example, when the two characteristic values ​​are m=2 and a=1, the corresponding number in Table 2 is 39. This means that the number of days the difference between the daily average current values ​​is continuously greater than 0 is 2, and the cumulative increase in the daily average current value over 2 consecutive days is 1. When both conditions are met, an alarm is triggered. There are 39 alarm devices, including faulty devices. When the number in Table 2 is 1, only one device alarms, and that device is faulty. Under these conditions, only faulty devices alarm, and non-faulty devices do not alarm, meaning the alarm is accurate and there are no false alarms.

[0044] 2) Trend 2: The current value rises slowly and steadily over a long period. The value of m ranges from 3 to 8 with a step size of 1, and the value of a ranges from 0.1 to 1.9 with a step size of 0.1. Each adjustment of m and a forms a combination, which is used to trigger an alarm. Under different combinations, the number of devices that trigger alarms is counted, as shown in Table 3.

[0045] Table 3

[0046]

[0047] The optimal alarm settings, selected based on the model training results, aim for accurate alarms with minimal false alarms or missed alarms; that is, faulty devices alarm while non-faulty devices do not. When multiple conditions meet the criteria, the stricter condition is chosen to reduce the likelihood of false alarms on the validation set. Increasing the alarm lead time can relax the duration requirement. Based on the training results, the optimal alarm settings are as follows:

[0048] 1) The current value rises significantly in the short term.

[0049] m=3, a=5, meaning the continuous rise lasted for 3 days, with a cumulative rise of 5 during that period.

[0050] 2) The current value rises slowly and steadily over a long period.

[0051] m=8, a=1.3, meaning the continuous rise lasted for 8 days, with a cumulative increase of 1.3 during that period.

[0052] Verification data: Current data of a sector section of a steel plant, selecting 121 current data points from 121 devices from January to April 2022.

[0053] Verification results:

[0054] 1) The current value rises significantly in the short term.

[0055] The alarm conditions selected for the training set were a continuous increase of 3 days and a cumulative increase of 5 during that period. Two characteristic values ​​were calculated for each device on the validation data. Two devices met the alarm conditions, but no abnormalities were reported on-site, and these were determined to be false alarms.

[0056] False alarms may be caused by early equipment failure, failure to detect abnormalities on-site, or increased current due to normal production operations. Although the false alarm rate is low, it is still recommended to check the equipment on-site to eliminate potential faults.

[0057] 2) The current value rises slowly and steadily over a long period.

[0058] The alarm condition selected for the training set was a continuous increase of 8 days with a cumulative increase of 1.3. On the validation data, two feature values ​​were calculated for each device; no alarms were generated, verifying the rationality of the model's feature parameter values.

[0059] This method has a wide range of applications. In addition to early warning of rising current trends, it has also verified the feasibility of early warning of vibration trends. It is particularly suitable for trend warning over a longer period of time. Compared with early warning of single-parameter over-limit, it can detect equipment failures in a timely manner and ensure the smooth operation of production.

Claims

1. A method for early warning of abnormalities in continuous casting equipment based on current change trends, characterized in that... This method includes the following steps: Step 1: Collect equipment fault information and collect current data from faulty and fault-free equipment as the selected training set; Step 2: Analyze the changing patterns of current data, filter out current values ​​in the shutdown state, and propose the mechanism for trend alarm; Step 3: If the current change trend is characterized by an increase in current value within a short period, the trend alarm mechanism will capture the current by controlling the cumulative increase in current within a short period. If the current change trend is characterized by a slow and stable increase in current value within a long period, the trend alarm mechanism will capture the current by controlling the number of consecutive days of increase and the cumulative increase in current during this period. Step 4: Set the over-limit condition characteristic value of the trend alarm mechanism: the difference of the daily average equipment current is >0 and the difference of the daily average equipment current within the consecutive m days is >a, where m is the number of consecutive days, a is the threshold of the difference of the daily average equipment current, and m and a are adjustable parameters. For devices where the current value increases within a short period, the value of m ranges from 2 to 3 with a step size of 1, and the value of a ranges from 1 to 9 with a step size of 1; for devices where the current value increases slowly and steadily within a long period, the value of m ranges from 3 to 8 with a step size of 1, and the value of a ranges from 0.1 to 1.9 with a step size of 0.

1. Step 5: Collect and calculate equipment current data, identify over-limit condition characteristic values, and provide an early warning of equipment anomalies.

2. The method for early warning of continuous casting equipment anomalies based on current change trends according to claim 1, characterized in that: Calculate the average current data of the equipment each day, and judge the trend of current change as rising or falling based on the change of the average value between two consecutive days. A difference of more than 0 between the daily average values ​​indicates an increase, and a difference of less than 0 indicates a decrease.

3. The method for early warning of continuous casting equipment anomalies based on current change trends according to claim 1, characterized in that: For equipment with a short-term current increase, set m=3 and a=5, meaning the current continues to rise for 3 days with a cumulative increase of 5. For equipment with a long-term current increase that is slow and stable, set m=8 and a=1.3, meaning the current continues to rise for 8 days with a cumulative increase of 1.3.

Images