Roller adaptive temperature measurement system and roller adaptive prediction algorithm

Abstract

The application discloses a kind of roller adaptive temperature measurement system and roller adaptive prediction algorithm, comprising: intelligent control module, internal integration machine learning algorithm, for predicting bearing failure and generating state prediction information and maintenance suggestion;Multi-sensor module is installed on roller, including temperature sensor, vibration sensor, pressure sensor;The multi-sensor module is used to collect roller operating condition data;Display panel is used to display the state prediction information and maintenance suggestion;Automatic alarm module includes sound alarm device and light alarm device;The automatic alarm module is used to remind personnel roller failure;Wireless transmission module is responsible for the signal connection between the multi-sensor module, intelligent control module, automatic alarm module and display panel;Therefore, by detecting bearing temperature, pressure and vibration, the possible failure of roller and the cause can be predicted, and suggestions can be generated;Therefore, the prompt accuracy of the existing monitoring system is effectively improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of metallurgy, and particularly relates to a rolling mill roll self-adaptive temperature measuring system and a rolling mill roll self-adaptive prediction algorithm. BACKGROUND

[0002] In the production of rods and wires in the metallurgical industry, the red steel is deformed due to the rotation of the rolling mill roll in the roughing stage. However, the roll in this stage has a high temperature of about 1000 DEG C. When the roll rotates, the bearing is easily burned due to the high temperature, improper oiling, bearing damage and other reasons. In turn, equipment failure and production stagnation are caused.

[0003] At present, the existing technology relies on manual temperature measurement or single temperature detection, and has the following problems:

[0004] 1. Hysteresis: only visual observation is used for alarm, and the fault cannot be predicted in advance;

[0005] 2. Singularity: lack of multi-dimensional data (such as vibration, pressure, etc.) support, high false alarm rate. SUMMARY

[0006] The technical problem to be solved by the application is to solve the above problems of the prior art, and to provide a rolling mill roll self-adaptive temperature measuring system and a rolling mill roll self-adaptive prediction algorithm. Through the detection of bearing temperature, pressure and vibration and other multi-dimensional data, the prediction algorithm is used to timely predict the possible faults of the rolling mill roll and the causes of the faults, and to generate solutions. Therefore, the application can effectively improve the prompt accuracy of the existing monitoring system and avoid false alarms.

[0007] To solve the above technical problems, the application adopts the following technical solutions:

[0008] A rolling mill roll self-adaptive temperature measuring system, comprising

[0009] An intelligent control module, which is internally integrated with a machine learning algorithm, is used to predict bearing failure and generate state prediction information and maintenance suggestions;

[0010] A multi-sensor module installed on the rolling mill roll, which comprises a temperature sensor, a vibration sensor and a pressure sensor, is used to collect rolling mill roll running state data in real time;

[0011] A display panel is used to display the state prediction information and maintenance suggestions;

[0012] An automatic alarm module comprising a sound alarm device and a light alarm device is used to remind personnel of the failure of the rolling mill roll;

[0013] A wireless transmission module is responsible for signal connection between the multi-sensor module, the intelligent control module, the automatic alarm module and the display panel.

[0014] The temperature sensor is annularly arranged three around the bearing center axis of each roller.

[0015] The wireless transmission module supports LoRa / Wi-Fi6 dual-mode communication.

[0016] S1, obtaining bearing temperature, bearing vibration frequency, bearing vibration amplitude and bearing load pressure value of roller state information, and preprocessing the roller state information;

[0017] S2, feature extraction is performed on the preprocessed roller state information, normal working condition data and fault working condition data of the roller are obtained, and the fault type in the fault working condition data is labeled;

[0018] S3, constructing a roller state prediction model to obtain a prediction result;

[0019] S4, based on the prediction result, comparing with the fault working condition data to judge the roller state, and generating maintenance suggestions combined with the fault type.

[0020] The process of generating maintenance suggestions in S4 includes: constructing a rule base, and matching the prediction result with the rule base to generate specific maintenance suggestions.

[0021] In S1, the preprocessing process of the roller state information includes the following steps:

[0022] S1.1, filtering the roller state information to remove noise in the sensor data;

[0023] S1.2, normalizing the obtained roller state information data to the same dimension in the range of 0-1;

[0024] S1.3, extracting temperature change rate, vibration frequency spectrum feature and pressure fluctuation amplitude from the original data.

[0025] The construction of the roller state prediction model includes the following steps:

[0026] T1, using LSTM neural network to construct time series prediction model, capturing the time sequence change rule of roller state through learning historical data;

[0027] T2, using random forest algorithm to classify and regress the feature data, and identifying the feature mode under different working conditions;

[0028] T3, evaluating the prediction effect of the model through cross-validation and model performance index, and adjusting and optimizing the model parameters;

[0029] T4, deploy the trained model to the roll adaptive temperature measurement system, receive new roll state data in real time, and perform state prediction and fault diagnosis.

[0030] In step T3, the prediction step of the roll state prediction model includes:

[0031] The pre-processed roll state data and the extracted feature data are input into the trained LSTM neural network and random forest model;

[0032] The LSTM neural network uses its internal memory unit and gating mechanism to calculate the input time series data, and outputs the roll health index and fault probability;

[0033] The random forest model classifies or regresses the feature data through the voting or averaging mechanism of multiple decision trees, and generates the corresponding prediction results;

[0034] The prediction results of the LSTM neural network and the random forest model are fused to obtain the final roll state prediction results, including the health index and the fault probability.

[0035] In step T4, the roll state prediction model determines the roll state, including:

[0036] The roll state prediction result is compared with the preset fault working condition data to analyze the similarity between the current roll state and the known fault mode;

[0037] According to the comparison result and the fault probability threshold, it is judged whether the roll is in normal working condition, warning state or fault state;

[0038] When the roll is determined to be in a fault state or the predicted fault probability is high, a hierarchical alarm mechanism is triggered to notify relevant personnel to take appropriate measures.

[0039] The present application has the following advantages:

[0040] The present application obtains multi-dimensional information such as bearing temperature, bearing vibration frequency, bearing vibration amplitude and bearing load pressure value through a multi-sensor module, obtains a prediction result through a roll state prediction model, judges the roll state based on the prediction result, timely predicts possible roll faults and the causes of the faults, and generates a solution suggestion. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 The present application provides a structure diagram of a roll adaptive temperature measurement system.

[0042] BRIEF DESCRIPTION OF DRAWINGS:1, display panel; 2, intelligent control module; 3, light alarm device; 4, temperature sensor; 5, roll. DETAILED DESCRIPTION

[0043] The application will be described in further detail below with reference to the drawings and specific preferred embodiments.

[0044] In the description of the application, it should be understood that the terms "left side", "right side", "upper part", "lower part" and the like indicate the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and "first", "second" and the like do not represent the importance of the parts, and therefore cannot be understood as limiting the application. The specific dimensions used in the embodiments are only for the purpose of illustrating the technical solutions and do not limit the protection scope of the application.

[0045] The application provides a roll self-adaptive temperature measurement system and a roll 5 self-adaptive prediction algorithm. Figure 1 , comprising an intelligent control module 2, which integrates machine learning algorithms inside, is used for predicting bearing failure and generating state prediction information and maintenance suggestions; a multi-sensor module installed on the roll 5, which comprises a temperature sensor 4, a vibration sensor and a pressure sensor; the multi-sensor module is used for collecting roll 5 running state data in real time; a display panel 1 for displaying the state prediction information and maintenance suggestions; an automatic alarm module comprising a sound alarm device and a light alarm device 3; the automatic alarm module is used for reminding personnel of roll 5 failure; a wireless transmission module responsible for signal connection between the multi-sensor module, the intelligent control module 2, the automatic alarm module and the display panel 1. Among them, the temperature sensor 4 monitors the bearing temperature in real time; the vibration sensor monitors the bearing vibration frequency and amplitude; the pressure sensor monitors the bearing load condition.

[0046] The temperature sensor 4 is arranged in a ring array around the bearing center axis of each roll 5.

[0047] The wireless transmission module supports LoRa / Wi-Fi 6 dual-mode communication.

[0048] The application also provides a roll 5 self-adaptive prediction algorithm, which is applied to the above-mentioned roll self-adaptive temperature measurement system, comprising the following steps:

[0049] Obtain roll 5 state information of bearing temperature, bearing vibration frequency, bearing vibration amplitude and bearing load pressure value, and preprocess the roll 5 state information;

[0050] Feature extraction is performed on the preprocessed state information of the roll 5 to obtain normal working condition data and fault working condition data of the roll 5, and the fault type in the fault working condition data is labeled;

[0051] A roll 5 state prediction model is constructed to obtain a prediction result;

[0052] Based on the prediction result, the roll 5 state is judged by comparing with the fault working condition data, and a maintenance suggestion is generated in combination with the fault type.

[0053] The process of generating the maintenance suggestion includes: constructing a rule base based on expert experience and historical data, and matching the prediction result with the rule base to generate a specific maintenance suggestion.

[0054] The preprocessing includes the following steps:

[0055] The roll 5 state information is filtered to remove noise in the sensor data;

[0056] The obtained roll 5 state information data is normalized to the same dimension in the range of 0-1;

[0057] The temperature change rate, vibration frequency spectrum features, and pressure fluctuation amplitude are extracted from the original data.

[0058] The construction of the roll state prediction model includes the following steps:

[0059] An LSTM neural network is used to construct a time series prediction model to capture the time sequence change rule of the roll state by learning the historical data;

[0060] The LSTM neural network model includes:

[0061] Input layer: the input data dimension is (T, D), where T is the time step and D is the feature number.

[0062] Hidden layer: contains multiple LSTM layers, and the number of hidden units in each layer is n. Wherein,

[0063] Input gate:

[0064] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0065] Forget gate:

[0066] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0067] Memory cell:

[0068]

[0069] Output gate:

[0070] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0071] Hidden state:

[0072] h t =o t ⊙tanh(C t )

[0073] where W i , W f , W C , W o are weight matrices, b i , b f , b C , b o are bias vectors, σ is the Sigmoid activation function, h t-1 is the hidden state of the previous time step, x t is the input feature of the current time step, C t-1 is the memory cell state of the previous time step, tanh is the hyperbolic tangent activation function, and ⊙ represents element-wise multiplication.

[0074] Output layer: outputs two nodes, one for predicting the roll health index and the other for predicting the roll failure probability. The calculation formula of the output layer is:

[0075]

[0076] where Dense represents a fully connected layer, and h T is the hidden state of the last time step.

[0077] Then use the random forest algorithm to classify and analyze the feature data, and identify the feature patterns under different working conditions;

[0078] where the random forest model includes:

[0079] Input features: the same feature vector as the LSTM model.

[0080] Decision tree: build multiple decision trees, and for the kth decision tree T k, input feature vectors x and target variables y, by recursively splitting the feature space. The splitting criterion for a decision tree is either Gini index or information gain. Among them,

[0081] Node Splitting: For each node j, select a feature vector x i and a threshold θ, split the data into two child nodes:

[0082] Node j = {x | x i ≤ θ} and Node j = {x | x i > θ}

[0083] where the selection of feature vector x i and threshold θ is based on Gini index or information gain.

[0084] Random Feature Selection:

[0085] When building each decision tree, randomly select m features from the feature set {x 1 , x 2 , x 3 , …, x D} for node splitting, where m ≤ D.

[0086] Voting or Averaging:

[0087] For classification tasks, determine the final prediction result by majority voting; specifically as follows:

[0088] For input x, each decision tree T k outputs a class prediction The final prediction result is the majority vote result:

[0089]

[0090] For regression tasks, determine the final prediction result by averaging; specifically as follows:

[0091] For input x, each decision tree T k outputs a numerical prediction The final prediction result is the average value:

[0092]

[0093] where:

[0094] K is the number of decision trees;

[0095] m is the number of features randomly selected by each decision tree for node splitting; for classification tasks, usually for regression tasks, usually

[0096] D is the total number of input features;

[0097] θ is the threshold for node splitting, used to divide data into two sub-nodes;

[0098] T k is the kth decision tree, used for prediction of input data;

[0099] Mode is the majority voting function, used for final prediction of classification tasks;

[0100] is the prediction result of the kth decision tree;

[0101] The prediction results of LSTM neural network and random forest algorithm are fused and combined output.

[0102] The prediction effect of the model is evaluated by cross-validation and model performance indicators, and the model parameters are adjusted and optimized;

[0103] The trained model is deployed in the roll adaptive temperature measurement system, which receives new roll 5 state data in real time, and performs state prediction and fault diagnosis.

[0104] When constructing the dataset in model training, collect historical data, including sensor data under normal and fault conditions; and, label the fault types (such as bearing wear, insufficient lubrication, temperature overrun, etc.).

[0105] When training the model, divide the historical data into training set and test set; then, use the training set to train the model and optimize the model parameters, such as the number of hidden nodes of LSTM and the depth of random forest; finally, use the test set to evaluate the model performance.

[0106] The prediction steps of the roll 5 state prediction model include:

[0107] Input the preprocessed roll 5 state data and extracted feature data into the trained LSTM neural network and random forest model;

[0108] The LSTM neural network uses its internal memory unit and gating mechanism to calculate the input time series data, and outputs the roll 5 health index and fault probability;

[0109] The random forest model uses voting or averaging mechanism of multiple decision trees to classify or regress the feature data, and generates the corresponding prediction results;

[0110] The prediction results of LSTM neural network and random forest model are fused to get the final roll 5 state prediction results, including health index and fault probability.

[0111] The step of determining the roller 5 state by the roller 5 state prediction model includes:

[0112] Comparing the roller 5 state prediction result with the preset fault condition data, analyzing the similarity between the current roller 5 state and the known fault mode;

[0113] According to the comparison result and the fault probability threshold, it is judged whether the roller 5 is in normal condition, early warning state or fault state;

[0114] When it is determined that the roller 5 is in a fault state or the predicted fault probability is high, a hierarchical alarm mechanism is triggered to inform relevant personnel to take appropriate measures.

[0115] Working principle:

[0116] Real-time data input;

[0117] Input the sensor data at the current time into the trained model;

[0118] The model outputs the prediction results, such as bearing health index and fault probability;

[0119] Fault diagnosis;

[0120] According to the prediction result, the bearing state is judged, such as normal, early warning and fault;

[0121] Combined with the fault type, maintenance suggestions are generated, such as increasing lubricating oil when the rack bearing temperature is too high.

[0122] Among them, the generation of maintenance suggestions is realized through the following steps:

[0123] Rule engine:

[0124] Based on expert experience and historical data, a rule base is constructed, such as classifying it as insufficient lubrication when the temperature is greater than 900℃ and the vibration is greater than 2kHz;

[0125] Match the model prediction result with the rule base to generate specific maintenance suggestions.

[0126] Dynamic optimization:

[0127] According to the actual maintenance effect feedback, dynamically adjust the rule base and model parameters to provide prediction accuracy.

Claims

1. A roll adaptive prediction algorithm, applied to a roll adaptive temperature measurement system, characterized in that, The adaptive temperature measurement system for the rolls includes: The intelligent control module (2) integrates machine learning algorithms to predict bearing failures and generate condition prediction information and maintenance suggestions. A multi-sensor module is installed on the roll (5), which includes a temperature sensor (4), a vibration sensor, and a pressure sensor; the multi-sensor module is used to collect real-time operating status data of the roll (5); Display panel (1) is used to display the status prediction information and maintenance suggestions; An automatic alarm module, which includes an audible alarm device and a visual alarm device (3); the automatic alarm module is used to remind personnel that the roller (5) has malfunctioned; The wireless transmission module is responsible for the signal connection between the multi-sensor module, the intelligent control module (2), the automatic alarm module and the display panel (1); The adaptive prediction algorithm for rolls includes the following steps: S1. Obtain the bearing temperature, bearing vibration frequency, bearing vibration amplitude and bearing load pressure value of the roll (5) and preprocess the bearing (5) status information; S2. Extract features from the preprocessed state information of the roll (5) to obtain normal working condition data and fault working condition data of the roll (5), and mark the fault type in the fault working condition data. S3. Construct a prediction model for the state of the roll (5) and obtain the prediction results; S4. Based on the prediction results, the state of the roll (5) is determined by comparing it with the fault condition data; and maintenance suggestions are generated in combination with the fault type.

2. The adaptive prediction algorithm for rolling mill rolls according to claim 1, characterized in that, The temperature sensors (4) are arranged in a ring array around the bearing center axis of each of the rolls (5).

3. The adaptive prediction algorithm for rolling mills according to claim 1, characterized in that, The wireless transmission module supports LoRa / Wi-Fi 6 dual-mode communication.

4. The adaptive prediction algorithm for rolling mills according to claim 1, characterized in that, The process of generating maintenance recommendations in S4 includes: building a rule base and matching the prediction results with the rule base to generate specific maintenance recommendations.

5. The adaptive prediction algorithm for rolling mills according to claim 1, characterized in that, In S1, the preprocessing of the state information of the roll (5) includes the following steps: S1.1 Filter the state information of the roll (5) to remove noise from the sensor data; S1.2 Normalize the obtained roll (5) status information data to the same dimension in the range of 0-1; S1.3 Extract temperature change rate, vibration spectrum characteristics, and pressure fluctuation amplitude from the raw data.

6. The adaptive prediction algorithm for rolling mill rolls according to claim 4, characterized in that, The steps for constructing a roll (5) state prediction model are as follows: T1. Use LSTM neural network to build time series prediction model, and capture the time series change law of roll (5) state by learning from historical data; T2. Use the random forest algorithm to classify and regress the feature data to identify feature patterns under different working conditions; T3. Evaluate the model's predictive performance through cross-validation and model performance metrics, and adjust and optimize the model parameters. T4. Deploy the trained model into the adaptive temperature measurement system of the roll (5) to receive new roll (5) status data in real time and perform status prediction and fault diagnosis.

7. The adaptive prediction algorithm for rolling mills according to claim 6, characterized in that, In step T4, the prediction steps of the roll (5) state prediction model include: Input the preprocessed roll (5) state data and extracted feature data into the trained LSTM neural network and random forest model; The LSTM neural network uses its internal memory units and gating mechanism to calculate the input time-series data and output the health index and failure probability of the roll (5). Random forest models use voting or averaging mechanisms among multiple decision trees to classify or regress feature data and generate corresponding prediction results. The prediction results of the LSTM neural network and the random forest model are fused to obtain the final roll (5) state prediction results, including the health index and failure probability.

8. The adaptive prediction algorithm for rolling mills according to claim 7, characterized in that, In step T4, the steps of the roll (5) state prediction model to determine the state of roll (5) include: The predicted state of the roll (5) is compared with the preset fault condition data to analyze the similarity between the current state of the roll (5) and the known fault modes. Based on the comparison results and the fault probability threshold, determine whether the roll (5) is in normal working condition, warning state or fault state; When the roll (5) is determined to be in a faulty state or the probability of a fault is predicted to be high, a graded alarm mechanism is triggered to notify relevant personnel to take corresponding measures.

Images