Base rod forming device and fault diagnosis method and system thereof

Abstract

The invention belongs to the technical field of fault diagnosis, and particularly relates to a base rod forming device and a fault diagnosis method and system thereof. The method comprises the following steps: S1, acquiring fault diagnosis section data in an operation process of a base rod forming device; the fault diagnosis data comprises real-time operation parameters and physical state parameters of the base rod forming device; the real-time operation parameters comprise the real-time operation speed of a pressing roller used for shredding sheets, the real-time operation speed of a cloth belt used for conveying forming paper and the real-time operation speed when a cutter cuts a base rod; the physical state parameters comprise one or a combination of two or three of temperature, pressure and vibration; and S2, inputting the fault diagnosis data into a pre-trained fault diagnosis model in the form of time sequence data to obtain the fault type of the base rod forming device. The technical problem that in the prior art, misjudgment and missed judgment are prone to occurring in fault diagnosis of the base rod forming device is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault diagnosis, and in particular relates to a base rod forming device and a fault diagnosis method and system thereof. Background Art

[0002] The primary component of a cigarette is the smoking section, which is typically made from thin tobacco paper with a certain amount of smoking agent added, then cut and rolled into shape. The base rod forming device is used to produce ordered thin, smoking sections. This process involves four steps: slitting the sheets, wrapping them with forming paper, gluing and shaping them, and cutting them to length, enabling mass production of heated cigarette base rods. A motor-driven roller rotates relative to each other, shearing the sheets into filaments. The sheets then pass through a forming and gathering mechanism, where they are wrapped with wrapping paper and evenly coated with glue. The resulting product is then dried at high temperature. Finally, the cut-to-length slitting device cuts the base rods into the desired length.

[0003] As a key manufacturing equipment in the field of cigarette production, the base rod forming device has a direct impact on production efficiency and economic benefits due to its operational stability and product quality. However, in the traditional production model, the fault diagnosis and production quality abnormality warning of the base rod forming device mainly rely on manual experience and simple statistical analysis, which has significant technical shortcomings. Operators usually conduct regular inspections, manually record equipment parameters, and make fault judgments based on historical data. However, during the base rod forming process, slight deviations in roundness and length may be caused by the coupling of multiple factors (such as mold wear, temperature fluctuations, material unevenness, etc.). Traditional methods are difficult to analyze such complex cause-and-effect relationships in real time, resulting in frequent misjudgments or missed judgments, which seriously affect product quality control.

[0004] Furthermore, the complexity of cigarette forming equipment further exacerbates the technical challenges. For example, a typical PROTOS-M5 cigarette making machine has over 2,000 moving parts, and a sensor network monitors dozens of parameters in real time, including temperature, air pressure, vibration, and motor current. Traditional fault diagnosis methods primarily rely on threshold alarm mechanisms, triggering an alarm when, for example, cigarette paper tension exceeds a set range. However, such methods lack sensitivity to progressive faults (such as rough cuts caused by blade wear) and are unable to distinguish between occasional noise interference and true fault signals. A study showed that in a case study at a Yunnan cigarette factory, the threshold alarm system had a false alarm rate of up to 65% for progressive mechanical faults, primarily due to a lack of effective exploitation of the temporal correlations in sensor data. For example, initial blade wear manifests itself as a slow increase in the vibration signal amplitude. Isolating the data at a single point in time can easily misinterpret this as random noise.

[0005] In summary, the fault diagnosis of the existing base rod forming device is prone to misjudgment and missed diagnosis problems. Summary of the Invention

[0006] The object of the present invention is to provide a base rod forming device and a fault diagnosis method and system thereof, so as to solve the technical problem in the prior art that fault diagnosis of the base rod forming device is prone to misjudgment and missed judgment.

[0007] To solve the above technical problems, the present invention provides a technical solution for a fault diagnosis method for a base rod forming device: a fault diagnosis method for a base rod forming device, when an abnormal production quality parameter is detected, a fault diagnosis method is used to determine the fault type of the base rod forming device; the fault diagnosis method includes:

[0008] S1. Acquiring fault diagnosis data during the operation of the base rod forming device; the fault diagnosis data includes real-time operating parameters and physical state parameters of the base rod forming device;

[0009] The real-time operating parameters include the real-time operating speed of the pressing roller for cutting the thin slices into shreds, the real-time operating speed of the cloth belt for conveying the forming paper, and the real-time operating speed of the cutter when cutting the base rod;

[0010] The physical state parameter includes one, or a combination of two or three of temperature, pressure and vibration;

[0011] S2. Input the fault diagnosis data into a pre-trained fault diagnosis model in the form of time series data to obtain the fault type of the base rod forming device.

[0012] The beneficial effect of the above technical solution is that the technical solution of the present invention, a method for diagnosing a base rod forming device fault, is a groundbreaking invention. Unlike the threshold alarm mechanism used in the prior art, the present invention provides a method for diagnosing a base rod forming device fault based on machine learning. Because various base rod forming device faults are closely related to the device's real-time operating parameters and physical state parameters, for example, abnormalities in the sheet slitting mechanism are related to the speed of the pressure roller that shreds the sheet, abnormalities in the tape conveyor mechanism are related to the tape speed, and abnormalities in the fixed-length cutting mechanism are related to the cutter speed, the fault type can be determined based on the fault diagnosis data. The present invention uses a fault diagnosis model to learn the inherent relationship between the device's fault diagnosis data and the fault type, achieving accurate fault diagnosis. The entire system performs excellently in production monitoring and equipment maintenance, providing an efficient and accurate technical solution for quality monitoring and equipment maintenance in industrial production. The present invention solves the technical problem of base rod forming device fault diagnosis being prone to misdiagnosis and missed detection in the prior art.

[0013] Furthermore, the real-time operating parameters in the fault diagnosis data also include one, or a combination of two, or three or more of the equipment operating speed, roller speed correction, belt speed correction, and belt loosening.

[0014] Furthermore, the fault diagnosis data also includes a cigarette detection signal, and the cigarette detection signal includes the length, roundness and appearance of the cut base rod.

[0015] Furthermore, the fault types include abnormal raw material status, abnormal sheet slitting mechanism status, abnormal tape conveying mechanism status and abnormal fixed-length slitting mechanism status.

[0016] Furthermore, the fault type includes one of the following: abnormal sheet roll state, abnormal slitting roller adjustment mechanism, abnormal slitting roller state, abnormal pulley transmission mechanism, abnormal motor reducer, abnormal cloth pulley mechanism state, abnormal tensioning pulley mechanism state, abnormal cutter transmission assembly and abnormal cutter power assembly, or a combination of two or three or more.

[0017] Furthermore, the fault diagnosis model includes a fault feature extraction module and a classification module; the fault feature extraction module is used to extract the time series features in the fault diagnosis segment data through LSTM, and the classification module is used to classify according to the time series features output by the fault feature extraction module to obtain the fault type.

[0018] Furthermore, the classification module includes a support vector machine.

[0019] Furthermore, the method for detecting abnormal production quality parameters includes:

[0020] (1) obtaining production prediction data of a base rod forming device during a base rod forming process; the production prediction data includes real-time operating parameters of the base rod forming device;

[0021] (2) inputting the production prediction data into a pre-trained production quality prediction model in the form of time series data to obtain production quality parameters of the base rod to be formed; the production quality parameters include base rod length and / or base rod roundness;

[0022] (3) Produce production quality warnings based on the production quality parameters obtained in (2).

[0023] Furthermore, the production prediction data also includes a cigarette detection signal, and the cigarette detection signal includes the length, roundness and appearance of the cut base rod.

[0024] Furthermore, the real-time operating parameters in the production forecast data also include equipment operating speed and / or set gluing speed.

[0025] Furthermore, the production quality prediction model includes a production feature extraction module and an attention mechanism module. The production feature extraction module is used to extract the time series features in the production prediction data, and the attention mechanism module is used to identify and weight the key features in the time series features output by the production feature extraction module to obtain production quality prediction parameters based on the weighted features.

[0026] Furthermore, the attention mechanism module is a hybrid attention mechanism including a temporal attention layer and a feature attention layer, and the hybrid attention mechanism includes a hybrid attention mechanism based on a parallel hybrid strategy or a hybrid attention mechanism based on a serial hybrid strategy;

[0027] The hybrid attention mechanism based on the parallel hybrid strategy is as follows: the temporal attention layer processes the temporal features output by the production feature extraction module through the self-attention mechanism to highlight important nodes in the temporal dimension; then, the feature attention layer processes the features output by the temporal attention layer through the channel attention mechanism to highlight important features in the feature dimension; finally, the output of the temporal attention layer and the output of the feature attention layer are combined through the gating mechanism;

[0028] The hybrid attention mechanism based on the serial hybrid strategy is as follows: the time attention layer processes the temporal features output by the production feature extraction module through the self-attention mechanism to highlight important nodes in the time dimension; then, the feature attention layer processes the features output by the time attention layer through the channel attention mechanism to highlight important features in the feature dimension.

[0029] Furthermore, the temporal attention layer is:

[0030]

[0031] Q=XW q

[0032] K=XW k

[0033] X t =A t ·X

[0034] Among them, X is the time series feature extracted by LSTM; At is the time attention weight matrix; W q 、W k are the learnable query weight matrix and key weight matrix respectively; X t It is the feature output by the temporal attention layer.

[0035] Furthermore, the feature attention layer is:

[0036] s=σ(W2×RELU(W1×GAP(Xt )))

[0037] X f =s⊙X t

[0038] Among them, X t is the feature output by the temporal attention layer; X f is the feature output by the feature attention layer; s is the feature attention weight vector; GAP represents global average pooling, W1 and W2 are the weight parameters of the fully connected layer; RELU(·) is the RELU activation function; σ(·) is the sigmoid activation function; ⊙ is the Hadamard product.

[0039] The present invention also provides a technical solution for a fault diagnosis system of a base rod forming device: a fault diagnosis system of a base rod forming device, comprising a processor, wherein the processor is used to execute a computer program to implement the steps of the fault diagnosis method of the base rod forming device as described above.

[0040] The present invention also provides a technical solution for a base rod forming device: a base rod forming device, comprising a fault diagnosis unit, the fault diagnosis unit comprising a processor, the processor being used to execute a computer program to implement the steps of the fault diagnosis method for the base rod forming device as described above. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 Schematic diagram of fault types in an embodiment of a fault diagnosis method for a base rod forming device of the present invention;

[0042] Figure 2 This is a schematic diagram of production quality early warning and equipment fault diagnosis in an embodiment of the fault diagnosis method for a base rod forming device of the present invention. DETAILED DESCRIPTION

[0043] Different from the threshold alarm mechanism used in the prior art, the present invention provides a machine learning-based fault diagnosis method for a base rod forming device. Because various base rod forming device faults are closely related to the device's real-time operating parameters and physical state parameters, for example, abnormalities in the sheet slitting mechanism are related to the speed of the pressure roller that cuts the sheet, abnormalities in the tape conveyor mechanism are related to the tape speed, and abnormalities in the fixed-length cutting mechanism are related to the cutter speed, the fault type can be determined based on fault diagnosis data. The present invention uses a fault diagnosis model to learn the inherent connection between the device's fault diagnosis data and the fault type, achieving accurate fault diagnosis. The entire system excels in production monitoring and equipment maintenance, providing an efficient and accurate technical solution for quality monitoring and equipment maintenance in industrial production. This present invention addresses the prior art problem of base rod forming device fault diagnosis being prone to misdiagnosis and missed detection.

[0044] Implementation method of fault diagnosis method for base rod forming device:

[0045] A method for diagnosing a fault of a base rod forming device, such as Figure 2 As shown, including production quality warning (i.e. Figure 2 Abnormal warning model on the left) and equipment fault diagnosis (i.e. Figure 2 The fault diagnosis model on the right) consists of two parts.

[0046] Production quality warning: Production quality warning is mainly used to predict the production quality of finished base rods during the base rod forming process to obtain production quality parameters, and to issue production quality warnings based on the prediction results.

[0047] Production quality parameters include base rod length and base rod roundness. Specifically, the method for predicting base rod length and base rod roundness in this embodiment includes collecting production prediction data of a base rod forming device during the base rod forming process, inputting the collected production prediction data into a pre-trained production quality prediction model, and thereby determining the length and roundness of the base rod to be formed.

[0048] The predicted production quality parameters are compared with the normal production status. Deviations from the normal production status and the duration of the deviations are used to issue production quality warnings, triggering corresponding warning signals. For example, if the length or roundness of the base rod exceeds the preset normal range multiple times in a row, the system will issue an early warning, alerting the operator to the possibility of a production quality issue and the need to adjust process parameters or check equipment status.

[0049] Production prediction data primarily includes roller pressure 1, roller pressure 2, roller gap 1, roller gap 2, base rod forming unit speed feedback, roller speed feedback, tape speed feedback, cutter speed feedback, set glue application speed, cigarette inspection signals (including cigarette length, roundness, and appearance), and the real-time and set heater temperatures. These production status parameters are input into the generative quality prediction model as time series data, i.e., [batch, time series length, number of features].

[0050] Among them, pressure roller pressure 1 and pressure roller pressure 2 refer to the pressure at two different positions of the cutting roller respectively; similarly, pressure roller gap 1 and pressure roller gap 2 refer to the gaps at two different positions of the cutting roller respectively.

[0051] It should be noted that the above-mentioned cigarettes refer to the base rods cut by the fixed-length cutting mechanism.

[0052] Specifically, the pressure roller speed feedback (i.e., the real-time operating speed of the pressure roller used to cut the thin slice into filaments) is the operating speed feedback value of the pressure roller in the base rod forming device. The pressure roller (i.e., the cutting roller) is a key component in the base rod forming process. The base rod forming device uses two relatively rotating (opposite rotation directions) pressure rollers to cut the thin slice located between the two pressure rollers into filaments. The speed of the pressure roller directly affects the forming quality and dimensional accuracy of the base rod. For example, if the pressure roller speed is too fast, the surface of the base rod may be uneven, while if the speed is too slow, it may affect production efficiency. The pressure roller speed feedback is obtained through real-time monitoring by sensors and is used to analyze whether the model is within the normal operating range.

[0053] Belt speed feedback (i.e., the real-time running speed of the belt used to transport the forming paper): The belt is the material conveyor belt used to wrap the base rod during the forming process. The belt wraps the forming paper and drives the forming paper to wrap the tobacco shreds in the forming mechanism to form the base rod. Its speed feedback value reflects the belt's operating status. The stability of the belt speed has a direct impact on the size and appearance of the base rod. For example, unstable belt speed may cause wrinkles or uneven tightness on the base rod surface. Belt speed feedback is also monitored by sensors and used by the model to determine whether the belt is operating normally.

[0054] Cutter speed feedback: Cutter speed feedback refers to the speed of the cutter when cutting the base rod (the speed of the cutter's movement in the direction of the base rod's transport). The accuracy of the cutter speed is crucial to the accuracy of the base rod's length. If the cutter speed is too fast or too slow, the base rod length will not meet the standard. Generally, the cutter speed should be consistent with the base rod's transport speed to ensure that the cutter and base rod are relatively stationary. Cutter speed feedback is monitored in real time by a sensor and used to analyze the model to ensure that the cutter is functioning properly.

[0055] Base rod forming unit operating speed: This is the operating speed of the entire base rod forming unit, reflecting the speed of the entire production process. The stability of the operating speed has a direct impact on production efficiency and product quality. For example, unstable operating speed may cause stalls or discontinuities during the base rod forming process, affecting product quality. Operating speed is monitored in real time by speed detection sensors, which are used by the model to determine whether the entire unit is operating normally.

[0056] Cigarette inspection signals: These are signals generated during the base rod forming process to check the quality of cigarettes (a finished product). Inspections include quality indicators such as length, roundness, and appearance (i.e., the presence of defects and related information). These signals are monitored in real time by inspection equipment and used to analyze product quality standards.

[0057] The production quality prediction model of this embodiment includes a long short-term memory (LSTM) neural network and a hybrid attention mechanism consisting of temporal attention and feature attention. The LSTM is used to extract time series features from the production status parameters input into the production quality prediction model. The hybrid attention mechanism combines temporal attention and feature attention to automatically identify and weight key time points and important features in the time series.

[0058] Specifically, the hybrid attention mechanism adopts the encoder structure as the basic framework, and the input is multivariate time series data (i.e., the output features of LSTM) (X∈R T×D )(T is the time step, D is the feature dimension). The hybrid attention mechanism is implemented by cascading the temporal attention layer, the feature attention layer and the hybrid strategy:

[0059] Temporal attention layer: focuses on key nodes in the time dimension, such as the time when an abnormal event occurs. The input is the original time series X, and the self-attention mechanism is used to calculate the correlation between time steps:

[0060]

[0061] Q=XW q

[0062] K=XW k

[0063] Where W q ,W k is a learnable parameter, A t ∈R T×T is the temporal attention weight matrix.

[0064] The temporal attention layer outputs the weighted temporal feature X t =A t ·X

[0065] Feature attention layer: focuses on important indicators in the feature dimension, including the criticality of sensor data in early warning. The input is the time series feature X output by the time attention layer t ; Dynamically assign feature weights through channel attention mechanism:

[0066] s=σ(W2×RELU(W1×GAP(X t )))

[0067] Among them, GAP represents global average pooling, W1, W2 are the weight parameters of the fully connected layer, and s∈R D is the feature weight vector. RELU(·) is the RELU activation function; σ(·) is the sigmoid activation function.

[0068] The feature attention layer outputs the weighted feature X f =s⊙X t . Where ⊙ is the Hadamard product.

[0069] That is, the temporal feature X output by the temporal attention layer t After pooling by global average pooling, it passes through a fully connected layer with a RELU activation function (i.e., the fully connected layer corresponding to W1) and a fully connected layer with a sigmoid activation function (i.e., the fully connected layer corresponding to W2) in turn to obtain the feature weight vector s.

[0070] Hybrid strategy: This embodiment adopts two hybrid strategies: parallel fusion and series fusion; in actual application, either of the two hybrid strategies can be adopted.

[0071] Parallel fusion: Combine the outputs of temporal attention and feature attention through a gating mechanism:

[0072] X hybrid =α·X t +(1-α)·X f

[0073] Among them, α is a learnable parameter, X hybrid is a hybrid feature of time and feature importance.

[0074] Tandem fusion: Apply temporal attention first, then apply feature attention to the result. That is, the feature attention layer outputs feature X f This is a mixed feature.

[0075] The mixed features are mapped through the fully connected layer to output the predicted production quality parameters, namely the roundness prediction value and length prediction value of the base rod.

[0076] The production quality prediction model above uses the predicted values ​​for the roundness and length of the base rod, along with the input operating parameters and test signals, to determine if there are any quality anomalies. If anomalies are found, an early warning signal is issued to alert the operator to take timely measures.

[0077] Equipment fault diagnosis (i.e., fault diagnosis method): Detecting an abnormality in base rod length or roundness indicates an equipment anomaly, and equipment fault diagnosis can be used to determine the fault type. Equipment fault diagnosis is primarily used to provide timely warnings of equipment failures during base rod forming device operation.

[0078] First, the equipment failures need to be sorted out to obtain different failure types, such as abnormal raw material status, abnormal sheet slitting mechanism status, etc.; then, different failure types need to be analyzed to determine the influencing factors of different failure types; after that, the parameters that need to be collected are determined based on different influencing factors.

[0079] Overall combing as Figure 1 As shown, the abnormal conditions of the base rod forming device are generally divided into 6 categories, namely, abnormal raw material condition, abnormal sheet slitting mechanism condition, abnormal forming mechanism condition, abnormal forming paper conveying mechanism condition, abnormal cloth tape conveying mechanism condition and abnormal fixed-length cutting mechanism condition.

[0080] Abnormal raw material conditions can be categorized as abnormal sheet roll conditions, abnormal wrapping paper conditions, and abnormal air shaft conditions. Abnormal sheet roll conditions are related to the pressure roller condition, and fuzzy judgment can be made based on the pressure roller condition, collecting data on pressure roller pressure and roller gap. Abnormal air shaft conditions can be indirectly determined by using data on mandrel expansion, mandrel contraction, and mandrel status. Abnormal wrapping paper conditions currently lack detection signals to determine abnormality.

[0081] Abnormal status of the sheet slitting mechanism can be divided into abnormality of the shredding roller adjustment mechanism, abnormality of the shredding roller status, abnormality of the pulley transmission mechanism, and abnormality of the motor reducer status. Abnormality of the shredding roller adjustment mechanism may be caused by abnormal status of components such as the threaded rod moving seat, abnormality of the shredding roller status may be caused by abnormal conditions such as damage or looseness of the upper and lower rollers, abnormality of the pulley transmission mechanism may be caused by abnormal transmission of the belt conveyor roller, and abnormality of the motor reducer status may be caused by a malfunction of the motor or reducer. Therefore, the above abnormal conditions can be fuzzy judged by collecting roller gap data, roller pressure data, equipment operating speed data, and roller speed data (i.e., roller speed feedback).

[0082] Abnormal molding mechanism status can be categorized as abnormal wire feeder status, glue gun status, or heater status. Wire feeder rotation abnormalities and glue gun status abnormalities caused by glue curing within the glue tank, hose, or gun are not yet detectable. Heater status abnormalities can be directly identified using the heater's temperature signal (either too high or too low).

[0083] The abnormal state of the forming paper conveying mechanism can be divided into abnormal state of the paper tray seat mechanism and abnormal state of the forming paper adjustment mechanism. There is no detection signal to judge both.

[0084] Abnormal conditions of the tape transmission mechanism can be divided into abnormal conditions of the tape pulley mechanism, abnormal conditions of the tensioning pulley mechanism and abnormal conditions of the adjusting pulley mechanism. Among them, the abnormal condition of the tape pulley mechanism may be caused by abnormalities of the transmission shaft coupling or faults of the motor reducer, and the abnormal condition of the tensioning pulley mechanism may be caused by the position deviation of the threaded rod moving seat. Therefore, the above abnormal conditions can be detected by collecting tape speed data, equipment operating speed data and tape status data (i.e. Figure 2 The abnormal state of the adjustment wheel mechanism is caused by the position deviation of the knob, and there is no detection signal to make a judgment.

[0085] Abnormalities in the fixed-length cutting mechanism can be categorized as abnormalities in the cutter assembly, cutter drive assembly, and cutter power assembly. Abnormalities in the cutter drive assembly may be caused by eccentric component wear or belt drive failure, while abnormalities in the cutter power assembly may be caused by faults in the motor reducer or other components. All of these abnormalities can be fuzzily identified by collecting cutter speed data, equipment operating speed data, and cigarette speed data. Cutter assembly abnormalities are caused by blade wear, and no detection signals are currently available to identify them.

[0086] For the above fault types that can be directly or indirectly judged, the fault type can be accurately judged by directly collecting the corresponding data and performing logical judgment. Figure 2 Shown in yellow on the upper right.

[0087] For the above-mentioned fault types that require fuzzy judgment, this embodiment proposes a fault diagnosis method: collect fault diagnosis data during the operation of the base rod forming device (including the real-time operating parameters and physical state parameters of the base rod forming device), and input the collected fault diagnosis data into a pre-trained fault diagnosis model in the form of time series data, thereby obtaining the equipment fault type (the above-mentioned fault type that requires fuzzy judgment) and the severity of the fault. Figure 2 Shown in the blue part on the lower right.

[0088] In this embodiment, the fault types output by the fault diagnosis model include abnormal sheet roll status, abnormal cutting roller adjustment mechanism, abnormal cutting roller status, abnormal pulley transmission mechanism, abnormal motor reducer, abnormal cloth pulley mechanism status, abnormal tensioning pulley mechanism status, abnormal cutter transmission assembly and abnormal cutter power assembly.

[0089] In other implementations, the fault type output by the fault diagnosis model can also be abnormal raw material status, abnormal sheet slitting mechanism status, abnormal tape conveyor mechanism status, and abnormal fixed-length slitting mechanism status, without further distinction. Accordingly, the data labels used in training the fault diagnosis model are the aforementioned fault types.

[0090] In other embodiments, the fault types output by the fault diagnosis model include any one, two, or a combination of three or more of the following: abnormal sheet roll status, abnormal slitting roller adjustment mechanism, abnormal slitting roller status, abnormal pulley transmission mechanism, abnormal motor reducer, abnormal cloth pulley mechanism status, abnormal tensioner mechanism status, abnormal cutter transmission assembly, and abnormal cutter power assembly. Only the most relevant types of faults are detected. For example, only faults in the fixed-length slitting mechanism, namely, abnormal cutter transmission assembly and abnormal cutter power assembly, are detected. This allows the model to focus on diagnosing one or more types of faults, thereby improving the ability to identify the corresponding fault types. Accordingly, the data labels used when training the fault diagnosis model are only the most relevant types of faults.

[0091] Operating status parameters primarily include roller pressure 1, roller pressure 2, roller gap 1, roller gap 2, base rod forming unit operating speed feedback, roller speed feedback, tape speed feedback, cutter speed feedback, roller speed correction (adjusting roller speed to ensure slice quality and size), tape speed correction (adjusting tape speed to ensure base rod size and appearance), tape release, cigarette detection signals (including cigarette speed, length, roundness, and appearance), and physical state parameters (including equipment pressure, temperature, and vibration). These operating status parameters are input into the fault diagnosis model as time series data, i.e., [batch, time series length, number of features].

[0092] Roller speed feedback: This is the same as the roller speed feedback in the production quality prediction model and is used to diagnose roller failures. For example, abnormal roller speed may indicate a roller motor failure or roller wear.

[0093] Tape speed feedback: This is the same as the tape speed feedback in the production quality prediction model and is used to diagnose whether there is a fault in the tape system. For example, abnormal tape speed may indicate tape slack or a fault in the tape drive system.

[0094] Cutter speed feedback: This is the same as the cutter speed feedback in the production quality prediction model and is used to diagnose whether there is a malfunction in the cutter system. For example, abnormal cutter speed may indicate a problem such as a cutter motor failure or cutter wear.

[0095] The operating speed of the base rod forming device is the same as the operating speed in the production quality prediction model. It is used to diagnose whether the entire device has a fault. For example, an abnormal operating speed may indicate a fault in the device's transmission system or motor.

[0096] Cigarette detection signal: This signal, similar to the cigarette detection signal in the production quality prediction model, is used to diagnose whether product quality anomalies are caused by equipment failure. For example, frequent quality anomaly signals may indicate a potential equipment failure.

[0097] Physical state parameters: These include equipment temperature, pressure, vibration, and other parameters. These parameters are monitored by sensors and used to diagnose equipment faults such as overheating, abnormal pressure, or excessive vibration.

[0098] The aforementioned operating status parameters vary in importance. Among them, feedback on roller speed, tape speed, cutter speed, and equipment operating status parameters (such as temperature, pressure, and vibration) are essential and crucial. These parameters are directly related to the core process of base rod forming and the critical operating conditions of the equipment, forming the basis for accurate early warning and fault diagnosis. For example, a slight change in roller speed can directly affect the quality of base rod forming, while the temperature and vibration parameters of the equipment are key indicators for determining the health of the equipment. Real-time monitoring and analysis of these parameters enables the system to detect potential problems in advance and issue timely warnings, thereby effectively avoiding production accidents.

[0099] In contrast, while the cigarette detection signal and the overall operating speed of the base rod forming unit contribute to the system's comprehensiveness and accuracy, they can be considered secondary parameters in some cases. The cigarette detection signal is primarily used for final inspection of finished product quality; its absence does not directly impact the system's ability to monitor the production process in real time and diagnose faults. While overall operating speed reflects production efficiency, it is not a core parameter in early warning and diagnostic models. Even without these parameters, the system can still effectively monitor and diagnose production quality and equipment status through other key parameters.

[0100] The fault diagnosis model of this embodiment includes LSTM and support vector machine (SVM) parts cascaded in sequence. LSTM is used to extract time series features from the operating status parameters input into the fault diagnosis model. SVM is used to replace the Softmax classifier as the final classification discriminator to output the type of fault. It can better handle the class imbalance problem between fault samples and normal samples, and significantly improve the accuracy and robustness of fault diagnosis. Through the effective extraction of time series features by the LSTM layer and the collaborative optimization of SVM in the early warning and diagnosis stages, the entire system performs well in production monitoring and equipment maintenance, providing an efficient and accurate technical solution for quality monitoring and equipment maintenance in industrial production.

[0101] The fault diagnosis model in this implementation accurately identifies the type of equipment failure. Once identified, it uses digital twin technology to provide root cause analysis and repair recommendations. When the system detects a fault of type "XX," the twin model automatically generates a repair work order and pushes it to the MES system, providing a detailed fault tree analysis and recommended replacement parts priorities. This predictive maintenance model not only extends equipment life but also reduces unplanned downtime, improving equipment reliability and maintenance efficiency.

[0102] This embodiment realizes accurate prediction of equipment failure and quality anomalies by constructing a time series data model, combining the improved IMV-LSTM (Interpretable Multi-Variable LSTM) production quality prediction model and the LSTM-SVM fault diagnosis model, and improves the efficiency of fault handling by linking the twin model through a visual interface. Specifically, the present invention proposes a complete set of intelligent monitoring solutions for common fault types (such as mold wear, sensor failure, pressure anomalies, etc.) and quality anomaly problems (such as base rod roundness deviation, length deviation, etc.) of the base rod forming device during the production process. The system realizes full-process automated management from data preprocessing, model training, fault diagnosis to decision response through multi-dimensional data acquisition, time series feature extraction, dynamic weight allocation and hybrid model optimization.

[0103] Implementation method of the fault diagnosis system of the base rod forming device:

[0104] A fault diagnosis system for a base rod forming device includes a processor configured to execute a computer program to implement the steps of the above-described method for diagnosing a fault of a base rod forming device. The specific method for diagnosing a fault of a base rod forming device has been described in sufficient detail in the above-described method for diagnosing a fault of a base rod forming device and will not be repeated here.

[0105] Specifically, the processor may be a CPU, or other general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor may also be a processor that supports the Advanced RISC Machine (ARM) architecture.

[0106] Implementation method of base rod forming device:

[0107] A base rod forming device includes a fault diagnosis unit, the fault diagnosis unit including a processor configured to execute a computer program to implement the steps of the above-described method for diagnosing a fault of a base rod forming device. The specific method for diagnosing a fault of a base rod forming device has been described in sufficient detail in the above-described embodiment of the method for diagnosing a fault of a base rod forming device and will not be repeated here.

[0108] Specifically, the processor may be a CPU, or other general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor may also be a processor that supports the Advanced RISC Machine (ARM) architecture.

[0109] The present invention has the following characteristics:

[0110] 1. By introducing a hybrid attention mechanism (feature weighting module and time weighting module) and optimizing classifier selection (replacing Softmax with SVM), this invention can more accurately capture key features and time points in time series data, significantly improving the accuracy of production quality anomaly warnings and equipment fault diagnosis. This combination of hybrid attention and SVM enables the model to perform well in handling complex failure modes, especially in small sample sizes and with class imbalance.

[0111] 2. By combining the attention mechanism with the SVM optimization objective (maximizing margin), the proposed model effectively reduces sensitivity to noisy data and outliers, thereby enhancing the model's robustness. This robustness is particularly important for addressing the data noise and equipment operational uncertainties that may arise in real industrial environments, ensuring the model's stable operation under complex operating conditions.

[0112] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments without inventive effort, or replace some of the technical features therein with equivalents. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A fault diagnosis method for a base rod forming device, characterized in that: When an abnormality in a production quality parameter is detected, a fault diagnosis method is used to diagnose the fault type of the base rod forming device; the fault diagnosis method includes: S1. Acquiring fault diagnosis data during the operation of the base rod forming device; the fault diagnosis data includes real-time operating parameters and physical state parameters of the base rod forming device; The real-time operating parameters include the real-time operating speed of the pressing roller for cutting the thin slices into shreds, the real-time operating speed of the cloth belt for conveying the forming paper, and the real-time operating speed of the cutter when cutting the base rod; The physical state parameter includes one, or a combination of two or three of temperature, pressure and vibration; S2. Input the fault diagnosis data into a pre-trained fault diagnosis model in the form of time series data to obtain the fault type of the base rod forming device.

2. The fault diagnosis method of the base rod forming device according to claim 1, characterized in that: The real-time operating parameters in the fault diagnosis data further include one, or a combination of two, or three or more of the equipment operating speed, roller speed correction, belt speed correction, and belt loosening.

3. The fault diagnosis method of the base rod forming device according to claim 1, characterized in that: The fault diagnosis data also includes a cigarette detection signal, and the cigarette detection signal includes the length, roundness and appearance of the cut base rod.

4. The fault diagnosis method of the base rod forming device according to claim 1, characterized in that: The fault types include abnormal raw material status, abnormal sheet slitting mechanism status, abnormal tape conveying mechanism status and abnormal fixed-length slitting mechanism status.

5. The fault diagnosis method of the base rod forming device according to claim 1 or 4, characterized in that: The fault types include abnormal sheet roll status, abnormal slitting roller adjustment mechanism, abnormal slitting roller status, abnormal pulley transmission mechanism, abnormal motor reducer, abnormal cloth pulley mechanism status, abnormal tensioning pulley mechanism status, abnormal cutter transmission assembly and abnormal cutter power assembly, or a combination of two or three or more.

6. The fault diagnosis method of the base rod forming device according to claim 1, characterized in that: The fault diagnosis model includes a fault feature extraction module and a classification module; the fault feature extraction module is used to extract the time series features in the fault diagnosis segment data through LSTM, and the classification module is used to classify according to the time series features output by the fault feature extraction module to obtain the fault type.

7. The fault diagnosis method of the base rod forming device according to claim 6, characterized in that: The classification module includes a support vector machine.

8. The fault diagnosis method of the base rod forming device according to claim 1, characterized in that: Methods for detecting anomalies in production quality parameters include: (1) obtaining production prediction data of a base rod forming device during a base rod forming process; the production prediction data includes real-time operating parameters of the base rod forming device; (2) inputting the production prediction data into a pre-trained production quality prediction model in the form of time series data to obtain production quality parameters of the base rod to be formed; the production quality parameters include base rod length and / or base rod roundness; (3) Provide production quality warning based on the production quality parameters obtained in (2).

9. The fault diagnosis method of the base rod forming device according to claim 8, characterized in that: The production prediction data also includes a cigarette detection signal, and the cigarette detection signal includes the length, roundness and appearance of the cut base rod.

10. The fault diagnosis method of the base rod forming device according to claim 8, characterized in that: The real-time operating parameters in the production forecast data also include equipment operating speed and / or set gluing speed.

11. The fault diagnosis method of the base rod forming device according to claim 8, characterized in that: The production quality prediction model includes a production feature extraction module and an attention mechanism module. The production feature extraction module is used to extract the time series features in the production prediction data, and the attention mechanism module is used to identify and weight the key features in the time series features output by the production feature extraction module to obtain production quality prediction parameters based on the weighted features.

12. The fault diagnosis method of the base rod forming device according to claim 11, characterized in that: The attention mechanism module is a hybrid attention mechanism including a temporal attention layer and a feature attention layer, and the hybrid attention mechanism includes a hybrid attention mechanism based on a parallel hybrid strategy or a hybrid attention mechanism based on a serial hybrid strategy; The hybrid attention mechanism based on the parallel hybrid strategy is as follows: the temporal attention layer processes the temporal features output by the production feature extraction module through the self-attention mechanism to highlight important nodes in the temporal dimension; then, the feature attention layer processes the features output by the temporal attention layer through the channel attention mechanism to highlight important features in the feature dimension; finally, the output of the temporal attention layer and the output of the feature attention layer are combined through the gating mechanism; The hybrid attention mechanism based on the serial hybrid strategy is as follows: the time attention layer processes the temporal features output by the production feature extraction module through the self-attention mechanism to highlight important nodes in the time dimension; then, the feature attention layer processes the features output by the time attention layer through the channel attention mechanism to highlight important features in the feature dimension.

13. The fault diagnosis method of the base rod forming device according to claim 12, characterized in that: The temporal attention layer is: Q=XW q K=XW k X t =A t ·X Among them, X is the time series feature extracted by LSTM; A t is the temporal attention weight matrix; W q 、W k are the learnable query weight matrix and key weight matrix respectively; X t It is the feature output by the temporal attention layer.

14. The fault diagnosis method of the base rod forming device according to claim 12, characterized in that: The feature attention layer is: s=σ(W2×RELU(W1×GAP(X t ))) X f =s⊙X t Among them, X t is the feature output by the temporal attention layer; X f is the feature output by the feature attention layer; s is the feature attention weight vector; GAP represents global average pooling, W1 and W2 are the weight parameters of the fully connected layer; RELU(·) is the RELU activation function; σ(·) is the sigmoid activation function; ⊙ is the Hadamard product.

15. A fault diagnosis system for a base rod forming device, comprising a processor, characterized in that: The processor is configured to execute a computer program to implement the steps of the fault diagnosis method for a base rod forming device according to any one of claims 1 to 14.

16. A base rod forming device, comprising a fault diagnosis unit, wherein the fault diagnosis unit comprises a processor, characterized in that: The processor is configured to execute a computer program to implement the steps of the fault diagnosis method for a base rod forming device according to any one of claims 1 to 14.

Images