Intelligent packaging production line control system and method

Abstract

The present invention belongs to the technical field of monitoring and fault warning of intelligent packaging production lines. The present invention discloses an intelligent packaging production line control system and method, which aims to achieve comprehensive monitoring and fault warning of the intelligent packaging production line; the control system comprises: collecting power parameter data of electrical components; the power parameter data includes current, voltage, energy utilization ratio data and vibration signals of the electrical components; preprocessing the power parameter data to form an electrical characteristic data set; training a fault state prediction model based on the electrical characteristic data set, and predicting the fault probability value of the electrical component based on the fault state prediction model; judging whether the intelligent packaging production line has a fault based on the fault probability value of the electrical component; when a fault is detected in the intelligent packaging production line, analyzing the faulty electrical component and taking targeted control measures to ensure efficient fault identification and control in various special environments.

Description

Technical Field

[0001] The present invention relates to the technical field of monitoring and fault warning of intelligent packaging production lines, and more specifically, to a control system and method for an intelligent packaging production line. Background Art

[0002] Intelligent packaging production line control systems are an integral part of modern industrial production. They not only enhance production automation but also ensure consistent and reliable product quality. With the advancement of technology, traditional packaging production lines are no longer able to meet the growing demand for efficiency, precision, and intelligence. Therefore, developing more advanced and efficient intelligent control systems has become a key area of ​​focus for the industry.

[0003] Existing intelligent packaging production line control systems perform poorly in fault detection under complex environments. Key issues include: data is susceptible to distortion due to electromagnetic interference and temperature fluctuations; difficulty integrating multi-source sensor data, which fails to fully reflect equipment status; reliance on static thresholds, which leads to false positives and false negatives; and a lack of real-time monitoring mechanisms, which delays problem resolution. Therefore, a more intelligent and comprehensive solution is urgently needed that ensures data accuracy, integrates multi-source information, and provides real-time, dynamic fault warning and control to improve production line reliability and efficiency. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned objectives, the present invention provides the following technical solutions: an intelligent packaging production line control system, comprising:

[0005] Power data acquisition module: used to collect power parameter data of electrical components; power parameter data includes current, voltage, energy utilization ratio data and vibration signals of electrical components;

[0006] Data processing module: used to pre-process power parameter data to form an electrical characteristic data set;

[0007] Abnormal analysis module: This module trains a fault status prediction model based on the electrical feature dataset and predicts the fault probability value of the electrical components based on the fault status prediction model.

[0008] Abnormal judgment module: used to judge whether there is a fault in the intelligent packaging production line based on the fault probability value of the electrical components;

[0009] The fault control module is used to analyze the faulty electrical components and take targeted control measures when a fault is detected in the intelligent packaging production line.

[0010] Furthermore, the method of preprocessing the power parameter data to form an electrical characteristic data set includes:

[0011] An improved temperature compensation algorithm is used to adjust current and voltage data, and the energy utilization ratio is calculated based on the adjusted current and voltage data. The vibration signal is converted into a spectrum using the Fourier transform method, and the spectrum is decomposed into a time series to extract signal features within different frequency ranges.

[0012] The adjusted current, voltage, energy utilization ratio data and signal characteristics in different frequency ranges are subjected to noise suppression, time alignment, standardization and data fusion processing to obtain an electrical characteristic data set;

[0013] The time alignment method is as follows: for each time period wt, the mean of current, voltage and energy utilization ratio at all times is calculated to obtain the power characteristics of the time period wt;

[0014] The data fusion method is to splice the standardized power characteristics and signal characteristics within the time period wt into a performance characteristic matrix by column, which is used as the electrical characteristic data set within the time period wt.

[0015] Furthermore, the method of adjusting current and voltage data using an improved temperature compensation algorithm includes:

[0016] Use thermistor sensor to measure the ambient temperature at any time t within the wt period, and use the improved temperature compensation algorithm to calculate the actual resistance value of the current sensor or voltage sensor at time t according to the ambient temperature. ;

[0017] in, represents the temperature of the current sensor or voltage sensor at time t, R( ) is the actual resistance value of the current sensor or voltage sensor at time t, Is the current sensor or voltage sensor at the reference temperature The resistance value is is the temperature coefficient;

[0018] Define room temperature as the reference temperature , measure the resistance value of the current sensor or voltage sensor at different temperatures, and record the corresponding temperature-resistance value ( , ),in, represents the rrth temperature value, Indicates the corresponding rrth resistance value;

[0019] The temperature-resistance value ( , ) is logarithmically transformed to linearize the exponential relationship ;

[0020] The logarithmic transformation formula is fitted using the least squares method, and we get and The best estimate of get ;

[0021] According to the actual resistance value, the current and voltage are adjusted; and the energy utilization ratio is calculated based on the adjusted current and voltage data.

[0022] Furthermore, the method of adjusting the current and voltage according to the actual resistance value includes:

[0023] For a shunt-based current sensor, the current correction factor of the current sensor at time t is calculated using the actual resistance value of the current sensor. , where DI represents the current sensor, represents the nominal resistance of the current sensor DI, Indicates the current correction coefficient of the current sensor, Indicates the actual resistance value of the current sensor;

[0024] Calculate the ratio of the current actually measured by the current sensor on the electrical component at time t to the correction coefficient to obtain the adjusted current data;

[0025] The electrical components include the main distribution cabinet, motor inverter, motor driver, central controller and heating elements in the intelligent packaging production line;

[0026] For a voltage sensor based on a voltage divider, the voltage correction factor of the voltage sensor at time t is calculated using the actual resistance value of the voltage sensor. , where DV represents the voltage sensor, is the voltage correction factor of the voltage sensor, and They are the two voltage divider resistors of the voltage sensor and The actual resistance value, and They are the two voltage divider resistors of the voltage sensor and Nominal resistance value;

[0027] The product of the voltage actually measured by the voltage sensor on the electrical component at time t and the voltage correction coefficient is calculated to obtain adjusted voltage data.

[0028] Furthermore, the method for calculating the energy utilization ratio using the adjusted current and voltage data includes:

[0029] Based on the adjusted current and voltage data, the product of current and voltage is calculated to obtain the apparent power SG;

[0030] Use a power analyzer to measure the power factor cos( ), calculate SG and cos( ) to obtain the active power ,in, represents the phase difference angle between current and voltage, Indicates power factor;

[0031] Calculate the apparent power and sin( ) to obtain the reactive power ,in, Represents the sinusoidal phase difference angle;

[0032] The ratio of active power to reactive power is taken to obtain the energy utilization ratio.

[0033] Furthermore, the method of converting the vibration signal into a frequency spectrum using the Fourier transform method includes:

[0034] The sampling rate is set according to the Nyquist sampling theorem, and the displacement sensor is used to collect the vibration signals of the electrical components of the intelligent packaging production line within the time period wt at the set sampling rate;

[0035] The collected vibration signal is filtered and the mean value is removed using bandpass filtering and averaging method;

[0036] Set a time window with a window width of wk;

[0037] Initialize the window index m to start from 0 and extract the signal segment of the m'th window from the processed vibration signal ZD[n] , where n represents the sample index in the window, which is used to traverse the sample points in each window, n=0,1,2,...,wk-1, and L represents the preset sliding step size, that is, the degree of overlap between adjacent windows;

[0038] Introduce a window function for the signal fragment within the window, use the window function as the weight of the signal fragment, define the window function as a Hamming window function, and apply it to the signal fragment within the window , and obtain the weighted signal segment , where hm[n] represents the Hamming window function, represents the signal segment after weighting in the mth time window;

[0039] For each weighted signal segment in the time window Apply fast Fourier transform to get the spectrum within the time window ,in, represents the spectrum of the mth time window at the Kth discrete frequency, K=0,1,2,...,wk-1, g is an imaginary unit, satisfying -1, e is a natural constant, It is a complex exponential function used to convert time domain signals to frequency domain;

[0040] The spectra in all time windows are integrated, with time as the horizontal axis and frequency as the vertical axis, to form a time-spectrum diagram containing the complete spectrum in the wt period.

[0041] Furthermore, the method of performing time series decomposition on the spectrum to extract signal features in different frequency ranges includes:

[0042] Based on the time-spectrum graph, all peak positions are marked in the time-spectrum graph, and for each peak, a frequency band including the peak and a preset ratio is selected as the frequency range;

[0043] For each frequency range, calculate the square of the corresponding spectrum value on the time-spectrum diagram and sum them to obtain the energy value of the frequency range, and select Ne frequency ranges with the largest energy value among all frequency ranges as frequency components;

[0044] Using Morlet wavelet as the basis function, a multi-scale continuous wavelet transform is performed for each time point in the time spectrum diagram to obtain the wavelet coefficient matrix at different scales. ;

[0045] in, represents the cth frequency component, c=1,2,...,Ne, represents a continuous time variable, represents the i-th scale, Represents the jth discrete position on the time axis of the time-frequency spectrum, j=1,2,...,J, x represents the spectral components in the time-frequency domain spectrogram, represents the complex conjugate of the Morlet wavelet function, Indicates 、 and The wavelet coefficients under ;

[0046] For each frequency component , build a scale To do The wavelet coefficient matrix with columns , all the calculated wavelet coefficients as the value for each position in the matrix;

[0047] According to the wavelet coefficient matrix under each frequency component, each scale Mapping to the corresponding frequency ,in, represents the sampling rate, represents the central angular frequency of the Morlet wavelet function;

[0048] Set the frequency band to low frequency band, middle frequency band and high frequency band, according to the frequency , all scales allocated to the corresponding frequency bands;

[0049] For each frequency band, calculate the sum of the energy of the wavelet coefficients of all relevant scales within it ,in, represents the total energy in the frequency band, represents all scales belonging to the frequency band, Indicates frequency band;

[0050] The energy of all frequency bands in the wavelet coefficient matrix is ​​combined into a vector to obtain the frequency band energy vector under each frequency component as the signal feature in each frequency range.

[0051] Furthermore, the training method of the fault state prediction model includes:

[0052] Collect sample sets, including the YB group performance data set and the corresponding true labels, and divide the sample sets into training sets and validation sets in proportion;

[0053] Construct a fault state prediction model and introduce L1 and L2 regularization into the model. The performance dataset is used as input and the fault probability value of the electrical component is used as the output label. The fault state prediction model is a gradient boosting decision tree model.

[0054] Initialize the model's hyperparameters and L1 regularization strength and L2 regularization strength ;

[0055] Define a containing multiple and The parameter network of the value is calculated, and the average F1 score of each parameter combination is calculated using the k-fold cross-validation method, and the parameter combination with the highest average F1 score is selected as the optimal parameter combination;

[0056] Define logarithmic loss as the loss function and introduce regularization terms, , where TL represents the loss function after the regularization term is introduced. represents the L1 regularization term, Represents the weight of the z-th feature, and takes the number of samples covered when the z-th feature is used as a split point in all trees as the weight of the feature, represents the L2 regularization term, Lo represents the logarithmic loss function;

[0057] The optimal parameter combination and the initialized model hyperparameters are used as model parameters, and the model parameters are adjusted according to the gradient in each iteration of the model using the training set and loss function;

[0058] For each iteration, the model parameters are updated using the training set, and the AUC value is calculated on the validation set using AUC as the evaluation indicator. A performance improvement threshold xnt is set. If the difference between the AUC value after the iteration and the AUC value of the previous iteration is greater than xnt, the model performance is judged to have improved. If the difference between the AUC value after the iteration and the AUC value of the previous iteration is less than or equal to xnt, the model performance is judged to have not improved. If the model performance on the validation set does not improve in consecutive pat iterations, the training is stopped. The trained gradient boosting decision tree model is obtained.

[0059] Furthermore, the method for determining whether a fault exists in the intelligent packaging production line based on the fault probability value of the electrical component includes:

[0060] Collect historical fault scores, sort them by time, and form a fault score sequence;

[0061] Set the sliding window HD and the sliding window step hd;

[0062] Apply the sliding window HD to the fault score sequence. For all fault scores in the initial sliding window, use the ROC curve method to calculate the true positive rate TPR and false positive rate FPR corresponding to each possible fault score threshold.

[0063] Based on the TPR and FPR within the initial sliding window, the optimal threshold yz is calculated using the Youden index;

[0064] Move the window forward by a step size nd, recalculate TPR and FPR for all fault scores in the new sliding window, and calculate the optimal threshold yz;

[0065] Take the average of all the best thresholds in the fault score sequence as the current best threshold;

[0066] When the fault score is greater than or equal to the optimal threshold, it means that the electrical components of the intelligent packaging production line are faulty. When the fault score is less than the optimal threshold, it means that the electrical components of the intelligent packaging production line are normal.

[0067] If any electrical component in the intelligent packaging production line fails, it is determined that the intelligent packaging production line has a fault.

[0068] Furthermore, when a fault is detected in the intelligent packaging production line, the method of analyzing the faulty electrical components and taking targeted control measures includes:

[0069] When a fault is detected on the intelligent packaging production line, the faulty electrical component type is analyzed and targeted control measures are taken for different types of electrical components, including: main distribution cabinet, motor inverter, motor driver, central controller and heating element;

[0070] For the main distribution cabinet, check and adjust the input voltage and current to within the preset normal range, optimize the load distribution within the main distribution cabinet to balance the load on each circuit, and check the setting values ​​of the circuit breakers to ensure that they can respond correctly to abnormal situations;

[0071] For motor inverters, we can obtain the actual status of the motor and adjust the motor's operating frequency to make the motor run in the best state. We can also optimize the start and stop procedures of the motor inverter to reduce the impact and downtime when the motor starts.

[0072] For motor drives, adjust the torque output of the motor drive, reduce the load on the motor drive, and optimize the speed setting of the motor to coordinate with other parts of the production line;

[0073] For the central controller, recalibrate the production parameters in the central controller to make the set production parameters meet the current production requirements;

[0074] For the heating element, check and adjust the temperature setting value of the heating element to make it within the preset safety range, and calibrate the temperature controller so that the accuracy of temperature measurement is greater than the preset accuracy threshold.

[0075] The technical effects and advantages of the intelligent packaging production line control system and method of the present invention are as follows:

[0076] The present invention aims to achieve comprehensive monitoring and fault warning of intelligent packaging production lines by integrating modules such as power data acquisition and adjustment, signal feature extraction, data processing, anomaly analysis and anomaly judgment; adjust current and voltage data using an improved temperature compensation algorithm, and calculate the energy utilization ratio based on the adjusted data, thereby improving the accuracy of current and voltage data and reducing the impact of ambient temperature changes on measurement results; collect vibration signals of electrical components, use Fourier transform to convert the vibration signals into frequency spectra, and perform time series decomposition to extract signal features within different frequency ranges, thereby achieving effective conversion from time domain to frequency domain, being able to capture vibration characteristics at different frequencies, providing detailed frequency domain information, and enhancing the ability to understand and analyze complex vibration patterns; preprocess power data and signal features, including noise suppression, time alignment, standardization and data fusion, to form a performance data set, eliminating Noise interference is eliminated, and time synchronization of multi-source data is achieved, which is convenient for comprehensive analysis; the improved gradient boosting decision tree (GBDT) algorithm is used in combination with L1 and L2 regularization to perform anomaly analysis on the performance data set, and the probability value of the fault state is output as the fault score, which effectively prevents model overfitting, improves the generalization ability and stability of the model, provides reliable fault prediction, and reduces false alarms and missed alarms; according to the fault score, the threshold is dynamically adjusted through the sliding window method, and the ROC curve and Youden index are used to determine the optimal threshold to judge whether there is a fault, realizing dynamic adjustment of the threshold, improving the sensitivity and specificity of fault detection, and reducing the false alarm rate and missed alarm rate, so that the system can not only detect and warn of potential faults in a timely manner, but also carry out targeted regulation of faulty electrical components, helping maintenance personnel to quickly locate and solve problems, thereby ensuring the efficient operation and stable production of the intelligent packaging production line. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 This is a schematic diagram of a control system for an intelligent packaging production line according to the present invention;

[0078] Figure 2 This is a schematic diagram of a control method for an intelligent packaging production line according to the present invention. DETAILED DESCRIPTION

[0079] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0080] Example 1:

[0081] See also Figure 1As shown, the intelligent packaging production line control system and method described in this embodiment include:

[0082] Power data acquisition module: used to collect power parameter data of electrical components; power parameter data includes current, voltage, energy utilization ratio data and vibration signals of electrical components;

[0083] Data processing module: used to pre-process power parameter data to form an electrical characteristic data set;

[0084] Abnormal analysis module: This module trains a fault status prediction model based on the electrical feature dataset and predicts the fault probability value of the electrical components based on the fault status prediction model.

[0085] Abnormal judgment module: used to judge whether there is a fault in the intelligent packaging production line based on the fault probability value of the electrical components;

[0086] The fault control module is used to analyze the faulty electrical components and take targeted control measures when a fault is detected in the intelligent packaging production line;

[0087] The method of preprocessing the power parameter data to form an electrical characteristic data set includes:

[0088] An improved temperature compensation algorithm is used to adjust current and voltage data, and the energy utilization ratio is calculated based on the adjusted current and voltage data. The vibration signal is converted into a spectrum using the Fourier transform method, and the spectrum is decomposed into a time series to extract signal features within different frequency ranges.

[0089] The adjusted current, voltage, energy utilization ratio data and signal characteristics in different frequency ranges are subjected to noise suppression, time alignment, standardization and data fusion processing to obtain an electrical characteristic data set;

[0090] The time alignment method is as follows: for each time period wt, the mean of current, voltage and energy utilization ratio at all times is calculated to obtain the power characteristics of the time period wt;

[0091] The data fusion method is to splice the standardized power characteristics and signal characteristics within the time period wt into a performance characteristic matrix by column, which is used as the electrical characteristic data set within the time period wt.

[0092] The method for adjusting current and voltage data using an improved temperature compensation algorithm includes:

[0093] Use a thermistor sensor to measure the ambient temperature at any time t within the wt period, and use the improved temperature compensation algorithm formula to calculate the actual resistance value of the current sensor (such as a shunt) or voltage sensor (such as a voltage divider) at time t based on the ambient temperature. ,in, represents the temperature of the current sensor or voltage sensor at time t, R( ) is the actual resistance value of the current sensor or voltage sensor at time t, Is the current sensor or voltage sensor at the reference temperature The resistance value under is the temperature coefficient;

[0094] Define room temperature as the reference temperature , measure the resistance value of the current sensor or voltage sensor at different temperatures, and record the corresponding temperature-resistance value ( , ),in, represents the rrth temperature value, Indicates the corresponding rrth resistance value;

[0095] The temperature-resistance value ( , ) is logarithmically transformed to linearize the exponential relationship ;

[0096] The logarithmic transformation formula is fitted using the least squares method, and we get and The best estimate of get ;

[0097] Adjust the current and voltage according to the actual resistance value, and calculate the power factor through the adjusted current and voltage data;

[0098] It should be noted that in highly automated packaging workshops, there are a large number of motors, inverters, and other electronic devices. The electromagnetic interference generated by these devices may affect the accuracy of power data. The working principle of most current sensors (such as Hall effect sensors or shunts) relies on changes in resistance. The increase in sensor temperature will cause the sensor's resistance to increase, thereby affecting the measured current value. Similarly, voltage sensors usually measure voltage through voltage dividers or other forms of resistor networks. Temperature changes will also affect the resistance of these resistor elements, thereby changing the measurement results. Therefore, after calculating the actual resistance value through the modified temperature compensation algorithm, this actual resistance value can be used to correct the output signals of the current and voltage sensors.

[0099] Improved temperature compensation algorithm ensures accurate current and voltage data even in complex electromagnetic environments. Real-time temperature sensor readings combined with temperature compensation dynamically adjust current and voltage data to eliminate the effects of temperature fluctuations and ensure data consistency and accuracy.

[0100] The method of adjusting the current and voltage according to the actual resistance value includes:

[0101] For a shunt-based current sensor, the current correction factor of the current sensor at time t is calculated using the actual resistance value of the current sensor. , where DI represents the current sensor, represents the nominal resistance of the current sensor DI, Indicates the current correction coefficient of the current sensor, Indicates the actual resistance value of the current sensor;

[0102] Calculate the ratio of the current actually measured by the current sensor on the electrical component at time t to the correction coefficient to obtain the adjusted current data;

[0103] The electrical components include the main power distribution cabinet (the main power supply of the production line), the motor inverter (the inverter used to control the speed and torque of the motor), the motor driver (the motor and its controller that drives the moving parts such as the conveyor belt and the robotic arm), the central controller (the central controller responsible for coordinating the operation of various subsystems), and the heating element (the device used to heat the sealing material).

[0104] For a voltage sensor based on a voltage divider, the voltage correction factor of the voltage sensor at time t is calculated using the actual resistance value of the voltage sensor. , where DV represents the voltage sensor, is the voltage correction coefficient of the voltage sensor, and They are the two voltage divider resistors of the voltage sensor and The actual resistance value, and They are the two voltage divider resistors of the voltage sensor and Nominal resistance value;

[0105] Calculate the product of the voltage actually measured by the voltage sensor on the electrical component at time t and the voltage correction coefficient to obtain adjusted voltage data;

[0106] It should be noted that the reason why voltage sensors have two resistors is usually related to the design of the voltage divider circuit. The voltage divider is a common component in voltage sensors that is used to convert a high voltage into a low voltage suitable for measurement. The voltage divider consists of two series resistors. These two resistors together determine how the input voltage is divided to the output terminal. The voltage divider is a simple circuit that uses two resistors. and To set the input voltage Proportionally reduced to the output voltage (i.e. the voltage actually measured by the sensor), its working principle is based on Ohm's law and Kirchhoff's voltage law, the formula is ;

[0107] The two resistors in the voltage sensor are mainly used to realize the voltage division function, ensuring that the output voltage is suitable for measurement. Proper selection of the resistance value can also realize the proportional adjustment, impedance matching, temperature compensation and protection functions. By performing temperature compensation on these two resistors, the accuracy of the measurement data can be further improved, ensuring the stable operation of the intelligent packaging production line.

[0108] The method for calculating the energy utilization ratio by using the adjusted current and voltage data includes:

[0109] Based on the adjusted current and voltage data, the product of current and voltage is calculated to obtain the apparent power SG;

[0110] Use a power analyzer to measure the power factor cos( ), calculate SG and cos( ) to obtain the active power ,in, represents the phase difference angle between current and voltage, Indicates power factor;

[0111] Calculate the apparent power and sin( ) to obtain the reactive power ,in, Represents the sinusoidal phase difference angle;

[0112] Take the ratio of active power to reactive power to obtain the energy utilization ratio PQ;

[0113] It should be noted that when the PQ value is high, it means that the proportion of active power relative to reactive power is large, which generally indicates that the system's energy utilization efficiency is high because more electrical energy is used to do useful work rather than being stored or released;

[0114] When the PQ value is low, it means that the reactive power is relatively large. This situation may indicate that there are more reactive loads in the system, such as induction motors or other inductive loads, which will lead to increased current, increased line losses, and may require larger cable and transformer capacities.

[0115] The method of converting the vibration signal into a frequency spectrum using the Fourier transform method comprises:

[0116] The sampling rate is set according to the Nyquist sampling theorem, and the displacement sensor is used to collect the vibration signals of the electrical components of the intelligent packaging production line within the time period wt at the set sampling rate;

[0117] The collected vibration signal is filtered and the mean value is removed using bandpass filtering and averaging method;

[0118] Set a time window with a window width of wk;

[0119] Initialize the window index m to start from 0 and extract the signal segment of the mth window from the processed vibration signal ZD[n] , where n represents the sample index in the window, used to traverse the sample points in each window, n=0,1,2,...,wk-1, and L represents the preset sliding step size, that is, the degree of overlap between adjacent windows, which is usually selected as a part of the window width (such as L=wt / 2 or L=wt / 4);

[0120] Introduce a window function for the signal fragment within the window, use the window function as the weight of the signal fragment, define the window function as a Hamming window function, and apply it to the signal fragment within the window , and obtain the weighted signal segment , where hm[n] represents the Hamming window function, represents the weighted signal segment in the mth time window;

[0121] It should be noted that the window function is a mathematical function used to weight the signal to reduce spectrum leakage and improve the quality of frequency domain analysis. It is applied to the signal segment within each window to generate a weighted signal.

[0122] For each weighted signal segment in the time window Apply fast Fourier transform to get the spectrum within the time window ,in, represents the spectrum of the mth time window at the Kth discrete frequency, K=0,1,2,...,wk-1, g is an imaginary unit, satisfying -1, e is a natural constant, It is a complex exponential function used to convert time domain signals to frequency domain;

[0123] Integrate the spectra in all time windows, with time as the horizontal axis and frequency as the vertical axis, to form a time-spectrum diagram containing the complete spectrum in the wt period;

[0124] It should be noted that in some environments with multivariable vibration sources, mechanical components in packaging production lines (such as motors and conveyor drives) will produce complex vibration patterns. These vibration patterns may mask potential fault signals. By converting the time domain vibration signal to the frequency domain, abnormal frequency components can be effectively identified.

[0125] The method of performing time series decomposition on the spectrum to extract signal features in different frequency ranges includes:

[0126] Based on the time-spectrum graph, all significant peak positions are marked in the time-spectrum graph. For each significant peak, a frequency band including the peak and a preset ratio (such as ±10% or ±20%) is selected as the frequency range;

[0127] For each frequency range, calculate the square of the corresponding spectrum value on the time-spectrum diagram and sum them to obtain the energy value of the frequency range, and select Ne frequency ranges with the largest energy value among all frequency ranges as frequency components;

[0128] Using Morlet wavelet as the basis function, a multi-scale continuous wavelet transform is performed for each time point in the time spectrum diagram to obtain the wavelet coefficient matrix at different scales. ;

[0129] in, represents the cth frequency component, c=1,2,...,Ne, Represents a continuous time variable, which is used to traverse the entire time axis and perform integration operations on the entire time axis to calculate the inner product between the wavelet basis function and the signal. In the integration process, it represents all possible time points, from negative infinity to positive infinity. Represents the i-th scale, which is obtained based on the frequency component, that is, , represents the sampling rate, represents the central angular frequency of the Morlet wavelet function (usually 5 or 6), is the displacement parameter, which represents the jth discrete position on the time axis of the time-frequency spectrum, j=1,2,...,J, and represents the specific time point of interest when calculating the wavelet coefficients, x Represents a spectral component in the time-frequency domain spectrogram, expressed as a function of time t, in The signal strength or amplitude under represents the complex conjugate of the Morlet wavelet function, Indicates 、 and The wavelet coefficients under ;

[0130] For each frequency component , build a scale To do The wavelet coefficient matrix with columns , all the calculated wavelet coefficients as the value for each position in the matrix;

[0131] According to the wavelet coefficient matrix under each frequency component, each scale Mapping to the corresponding frequency ,in, represents the sampling rate, represents the central angular frequency of the Morlet wavelet function;

[0132] Set the frequency band to low, middle and high (for example, set the frequency band to low 0-30Hz, middle 30-100Hz and high 100-300Hz), according to the frequency , all scales Assign to the corresponding frequency band (for example, if the frequency corresponding to a certain scale falls into the low frequency band, it will be classified into the low frequency band);

[0133] For each frequency band, calculate the sum of the energy of the wavelet coefficients of all relevant scales within it ,in, represents the total energy in the frequency band, represents all scales belonging to the frequency band, Indicates frequency band;

[0134] The energy of all frequency bands in the wavelet coefficient matrix is ​​combined into a vector to obtain the frequency band energy vector under each frequency component as the signal feature in each frequency range;

[0135] The training method of the fault state prediction model includes:

[0136] Collect sample sets, including the YB group performance data set and the corresponding true labels, and divide the sample sets into training sets and validation sets in proportion;

[0137] Construct a fault state prediction model and introduce L1 and L2 regularization into the model. The performance dataset is used as input and the fault probability value of the electrical component is used as the output label. The fault state prediction model is a gradient boosting decision tree model.

[0138] Initialize the model's hyperparameters and L1 regularization strength and L2 regularization strength ;

[0139] Define a containing multiple and The parameter network of the value is calculated, and the average F1 score of each parameter combination is calculated using the k-fold cross-validation method, and the parameter combination with the highest average F1 score is selected as the optimal parameter combination;

[0140] Define logarithmic loss as the loss function and introduce regularization terms, , where TL represents the loss function after the regularization term is introduced. represents the L1 regularization term, Indicates the regularization strength, controls the degree of regularization, and the larger the It will increase sparsity and may remove more features. Represents the weight of the z-th feature, and takes the number of samples covered when the z-th feature is used as a split point in all trees as the weight of the feature, represents the L2 regularization term, Indicates the L2 regularization strength, controls the degree of regularization, and the larger the It will make the model smoother and reduce the risk of overfitting. Lo represents the logarithmic loss function. , where YB represents the number of samples, represents the true label of the i-th sample, represents the predicted probability that the i-th sample is in a fault state;

[0141] The optimal parameter combination and the initialized model hyperparameters are used as model parameters, and the model parameters are adjusted according to the gradient in each iteration of the model using the training set and the loss function (the regularization term directly affects the gradient calculation, thereby controlling the parameter update);

[0142] For each iteration, the model parameters are updated using the training set, and the AUC value is calculated on the validation set using AUC as the evaluation metric;

[0143] Set a performance improvement threshold xnt. If the difference between the AUC value after the iteration and the AUC value of the previous iteration is greater than xnt, the model performance is judged to have improved. If the difference between the AUC value after the iteration and the AUC value of the previous iteration is less than or equal to xnt, the model performance is judged to have not improved. If the model performance on the validation set does not improve in consecutive pat iterations, stop training. Obtain a trained gradient boosting decision tree model.

[0144] It should be noted that introducing regularization terms into the gradient boosting decision tree can significantly improve model performance and technical effects. By adding regularization terms to the loss function, regularization effectively prevents overfitting, limits the size or number of model parameters, and reduces model complexity, making the model more general and performing better on unseen data. Regularization improves generalization ability, allowing the model to perform well not only on training data, but also on validation and test sets. In particular, when using L1 regularization, it encourages sparse solutions, reduces the influence of unimportant features, and enhances model interpretability. L2 regularization prevents parameters from being too large, making the model smoother and more stable, helping to stabilize the model training process and reduce the problem of gradient explosion or vanishing, especially in high-dimensional data sets. In addition, regularization can also accelerate model convergence, simplify feature selection, and improve the reliability and stability of the model in practical applications, reducing the risk of false positives and false negatives. In short, regularization makes the model more robust and efficient when processing complex data, improving overall performance and reliability.

[0145] In some environments that require long-term continuous operation, smart packaging production lines often need to run continuously for a long time, which increases the risk of electrical system failure. By improving the gradient boosting decision tree (GBDT), it can efficiently process large amounts of historical data, capture long-term dependencies, and predict future failure trends;

[0146] The method for determining whether a fault exists in the intelligent packaging production line based on the fault probability value of the electrical component includes:

[0147] Collect historical fault scores, sort them by time, and form a fault score sequence;

[0148] Set the sliding window HD and the sliding window step hd (usually set to half the time window size);

[0149] Apply the sliding window HD to the fault score sequence. For all fault scores in the initial sliding window, use the ROC curve method to calculate the true positive rate TPR and false positive rate FPR corresponding to each possible fault score threshold.

[0150] Based on the TPR and FPR within the initial sliding window, the optimal threshold yz is calculated using the Youden index;

[0151] Move the window forward by a step size nd, recalculate TPR and FPR for all fault scores in the new sliding window, and calculate the optimal threshold yz;

[0152] Take the average of all the best thresholds in the fault score sequence as the current best threshold;

[0153] When the fault score is greater than or equal to the optimal threshold, it means that the electrical components of the intelligent packaging production line are faulty. When the fault score is less than the optimal threshold, it means that the electrical components of the intelligent packaging production line are normal.

[0154] If any electrical component in the intelligent packaging production line fails, it is determined that the intelligent packaging production line has a fault;

[0155] When a fault is detected in the intelligent packaging production line, the method of analyzing the faulty electrical components and taking targeted control measures includes:

[0156] When it is determined that there is a fault in the intelligent packaging production line, the faulty electrical component category is analyzed and targeted control measures are taken for different categories of electrical components;

[0157] Main distribution cabinet, motor inverter, motor drive, central controller and heating elements;

[0158] For the main distribution cabinet, check and adjust the input voltage and current to within the preset normal range, optimize the load distribution within the main distribution cabinet to balance the load on each circuit, and check the setting values ​​of the circuit breakers to ensure that they can respond correctly to abnormal situations;

[0159] For motor inverters, we can obtain the actual status of the motor and adjust the motor's operating frequency to make the motor run in the best state. We can also optimize the start and stop procedures of the motor inverter to reduce the impact and downtime when the motor starts.

[0160] For motor drives, adjust the torque output of the motor drive, reduce the load on the motor drive, and optimize the speed setting of the motor to coordinate with other parts of the production line;

[0161] For the central controller, recalibrate the production parameters in the central controller to make the set production parameters meet the current production requirements;

[0162] For the heating element, check and adjust the temperature setting value of the heating element to make it within the preset safety range, and calibrate the temperature controller so that the accuracy of temperature measurement is greater than the preset accuracy threshold.

[0163] This embodiment realizes comprehensive monitoring and fault warning of the intelligent packaging production line by integrating modules such as power data acquisition and adjustment, signal feature extraction, data processing, anomaly analysis and anomaly judgment; uses an improved temperature compensation algorithm to adjust current and voltage data, and calculates the energy utilization ratio based on the adjusted data, thereby improving the accuracy of current and voltage data and reducing the impact of ambient temperature changes on measurement results; collects vibration signals of electrical components, uses Fourier transform method to convert vibration signals into frequency spectrum, and performs time series decomposition to extract signal features in different frequency ranges, realizing effective conversion from time domain to frequency domain, capable of capturing vibration characteristics at different frequencies, providing detailed frequency domain information, and enhancing the understanding and analysis capabilities of complex vibration patterns; preprocesses power data and signal features, including noise suppression, time alignment, standardization and data fusion, to form a performance data set, eliminates noise interference, realizes time synchronization of multi-source data, and facilitates comprehensive analysis; uses an improved gradient boosting decision tree The GBDT algorithm combines L1 and L2 regularization to perform anomaly analysis on the performance data set and outputs the probability value of the fault state as the fault score, which effectively prevents model overfitting, improves the generalization ability and stability of the model, provides reliable fault prediction, and reduces false positives and missed negatives. Based on the fault score, the threshold is dynamically adjusted through the sliding window method, and the ROC curve and Youden index are used to determine the optimal threshold to determine whether a fault exists. This achieves dynamic adjustment of the threshold, improves the sensitivity and specificity of fault detection, and reduces the false positive and missed negative rates. It can not only detect and warn of potential faults in a timely manner, but also provide detailed fault analysis reports to help maintenance personnel quickly locate and solve problems, thereby ensuring the efficient operation and stable production of smart packaging production lines. This method is particularly suitable for smart packaging production lines that have high stability requirements, complex environmental conditions, and require continuous and efficient operation. Through targeted technical means and innovative methods, it ensures efficient fault identification and preventive maintenance in various special environments.

[0164] Example 2:

[0165] See also Figure 2 As shown, for the parts not described in detail in this embodiment, please refer to the description of Example 1. A method for controlling an intelligent packaging production line is provided, including:

[0166] S1. Collecting power parameter data of electrical components; the power parameter data includes current, voltage, energy utilization ratio data and vibration signals of electrical components;

[0167] S2. Preprocessing the power parameter data to form an electrical characteristic data set;

[0168] S3. Obtain a fault state prediction model based on the electrical characteristic data set training, and obtain a fault probability value of the electrical component based on the fault state prediction model prediction;

[0169] S4. Determine whether the intelligent packaging production line has a fault based on the fault probability value of the electrical component;

[0170] S5. When a fault is detected in the intelligent packaging production line, analyze the faulty electrical components and take targeted control measures.

[0171] Example 3:

[0172] This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the intelligent packaging production line control method provided above is implemented.

[0173] Since the electronic device described in this embodiment is used to implement the intelligent packaging production line control method in the embodiments of this application, those skilled in the art will be able to understand the specific implementation and various variations of the electronic device in this embodiment based on the intelligent packaging production line control method described in the embodiments of this application. Therefore, how the electronic device implements the method in the embodiments of this application will not be described in detail here. As long as those skilled in the art can implement the electronic device used in the intelligent packaging production line control method in the embodiments of this application, it falls within the scope of protection of this application.

[0174] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0175] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that for users of ordinary skill in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. An intelligent packaging production line control system, characterized in that: include: Power data acquisition module: used to collect power parameter data of electrical components; Power parameter data includes current, voltage, energy utilization rate data and vibration signals of electrical components; Data processing module: used to pre-process power parameter data to form an electrical characteristic data set; Adjusting current and voltage data using an improved temperature compensation algorithm, and calculating an energy utilization ratio based on the adjusted current and voltage data; The vibration signal is converted into a spectrum using the Fourier transform method, and the spectrum is decomposed into a time series to extract signal features in different frequency ranges; The adjusted current, voltage, energy utilization ratio data and signal characteristics in different frequency ranges are subjected to noise suppression, time alignment, standardization and data fusion processing to obtain an electrical characteristic data set; The time alignment method is as follows: for each time period wt, the mean of current, voltage and energy utilization ratio at all times is calculated to obtain the power characteristics of the time period wt; The data fusion method is as follows: the standardized power characteristics and signal characteristics in the time period wt are spliced ​​into a performance characteristic matrix by column, which is used as the electrical characteristic data set in the time period wt; Methods for adjusting current and voltage data using an improved temperature compensation algorithm include: Use thermistor sensor to measure the ambient temperature at any time t within the wt period, and use the improved temperature compensation algorithm to calculate the actual resistance value of the current sensor or voltage sensor at time t according to the ambient temperature. ; in, represents the temperature of the current sensor or voltage sensor at time t, R( ) is the actual resistance value of the current sensor or voltage sensor at time t, Is the current sensor or voltage sensor at the reference temperature The resistance value is is the temperature coefficient; Define room temperature as the reference temperature , measure the resistance value of the current sensor or voltage sensor at different temperatures, and record the corresponding temperature-resistance value ( , ),in, represents the rrth temperature value, Indicates the corresponding rrth resistance value; The temperature-resistance value ( , ) is logarithmically transformed to linearize the exponential relationship ; The logarithmic transformation formula is fitted using the least squares method, and we get and The best estimate of get ; Adjust the current and voltage according to the actual resistance value; and calculate the energy utilization ratio through the adjusted current and voltage data; Abnormal analysis module: This module trains a fault status prediction model based on the electrical feature dataset and predicts the fault probability value of the electrical components based on the fault status prediction model. Abnormal judgment module: used to judge whether there is a fault in the intelligent packaging production line based on the fault probability value of the electrical components; The fault control module is used to analyze the faulty electrical components and take targeted control measures when a fault is detected in the intelligent packaging production line.

2. The intelligent packaging production line control system according to claim 1, characterized in that: The method of adjusting the current and voltage according to the actual resistance value includes: For a shunt-based current sensor, the current correction factor of the current sensor at time t is calculated using the actual resistance value of the current sensor. , where DI represents the current sensor, represents the nominal resistance of the current sensor DI, Indicates the current correction coefficient of the current sensor, Indicates the actual resistance value of the current sensor; Calculate the ratio of the current actually measured by the current sensor on the electrical component at time t to the correction coefficient to obtain the adjusted current data; The electrical components include the main distribution cabinet, motor inverter, motor driver, central controller and heating elements in the intelligent packaging production line; For a voltage sensor based on a voltage divider, the voltage correction factor of the voltage sensor at time t is calculated using the actual resistance value of the voltage sensor. , where DV represents the voltage sensor, is the voltage correction coefficient of the voltage sensor, and They are the two voltage divider resistors of the voltage sensor and The actual resistance value, and They are the two voltage divider resistors of the voltage sensor and Nominal resistance value; The product of the voltage actually measured by the voltage sensor on the electrical component at time t and the voltage correction coefficient is calculated to obtain adjusted voltage data.

3. The intelligent packaging production line control system according to claim 2, characterized in that: The method for calculating the energy utilization ratio by using the adjusted current and voltage data includes: Based on the adjusted current and voltage data, the product of current and voltage is calculated to obtain the apparent power SG; Use a power analyzer to measure the power factor cos( ), calculate SG and cos( ) to obtain the active power ,in, represents the phase difference angle between current and voltage, Indicates power factor; Calculate the apparent power and sin( ) to obtain the reactive power ,in, Represents the sinusoidal phase difference angle; The ratio of active power to reactive power is taken to obtain the energy utilization ratio.

4. The intelligent packaging production line control system according to claim 3, characterized in that: The method of converting the vibration signal into a frequency spectrum using the Fourier transform method comprises: The sampling rate is set according to the Nyquist sampling theorem, and the displacement sensor is used to collect the vibration signals of the electrical components of the intelligent packaging production line within the time period wt at the set sampling rate; The collected vibration signal is filtered and the mean value is removed using bandpass filtering and averaging method; Set a time window with a window width of wk; Initialize the window index m to start from 0 and extract the signal segment of the m'th window from the processed vibration signal ZD[n] , where n represents the sample index in the window, which is used to traverse the sample points in each window, n=0,1,2,...,wk-1, and L represents the preset sliding step size, that is, the degree of overlap between adjacent windows; Introduce a window function for the signal fragment within the window, use the window function as the weight of the signal fragment, define the window function as a Hamming window function, and apply it to the signal fragment within the window , and obtain the weighted signal segment , where hm[n] represents the Hamming window function, represents the weighted signal segment in the mth time window; For each weighted signal segment in the time window Apply fast Fourier transform to get the spectrum within the time window ,in, represents the spectrum of the mth time window at the Kth discrete frequency, K=0,1,2,...,wk-1, g is an imaginary unit, satisfying -1, e is a natural constant, It is a complex exponential function used to convert time domain signals to frequency domain; The spectra in all time windows are integrated, with time as the horizontal axis and frequency as the vertical axis, to form a time-spectrum diagram containing the complete spectrum in the wt period.

5. The intelligent packaging production line control system according to claim 4, characterized in that: The method of performing time series decomposition on the spectrum to extract signal features in different frequency ranges includes: Based on the time-spectrum graph, all peak positions are marked in the time-spectrum graph, and for each peak, a frequency band including the peak and a preset ratio is selected as the frequency range; For each frequency range, calculate the square of the corresponding spectrum value on the time-spectrum diagram and sum them to obtain the energy value of the frequency range, and select Ne frequency ranges with the largest energy value among all frequency ranges as frequency components; Using Morlet wavelet as the basis function, a multi-scale continuous wavelet transform is performed for each time point in the time spectrum diagram to obtain the wavelet coefficient matrix at different scales. ; in, represents the cth frequency component, c=1,2,...,Ne, represents a continuous time variable, represents the i-th scale, Represents the jth discrete position on the time axis of the time-frequency spectrum, j=1,2,...,J, x represents the spectral components in the time-frequency domain spectrogram, represents the complex conjugate of the Morlet wavelet function, Indicates 、 and The wavelet coefficients under ; For each frequency component , build a scale To do The wavelet coefficient matrix with columns , all the calculated wavelet coefficients as the value for each position in the matrix; According to the wavelet coefficient matrix under each frequency component, each scale Mapping to the corresponding frequency ,in, represents the sampling rate, represents the central angular frequency of the Morlet wavelet function; Set the frequency band to low frequency band, middle frequency band and high frequency band, according to the frequency , all scales allocated to the corresponding frequency bands; For each frequency band, calculate the sum of the energy of the wavelet coefficients of all relevant scales within it ,in, represents the total energy in the frequency band, represents all scales belonging to the frequency band, Indicates frequency band; The energy of all frequency bands in the wavelet coefficient matrix is ​​combined into a vector to obtain the frequency band energy vector under each frequency component as the signal feature in each frequency range.

6. The intelligent packaging production line control system according to claim 5, characterized in that: The training method of the fault state prediction model includes: Collect sample sets, including the YB group performance data set and the corresponding true labels, and divide the sample sets into training sets and validation sets in proportion; Construct a fault state prediction model and introduce L1 and L2 regularization into the model. The performance dataset is used as input and the fault probability value of the electrical component is used as the output label. The fault state prediction model is a gradient boosting decision tree model. Initialize the model's hyperparameters and L1 regularization strength and L2 regularization strength ; Define a containing multiple and The parameter network of the value is calculated, and the average F1 score of each parameter combination is calculated using the k-fold cross-validation method, and the parameter combination with the highest average F1 score is selected as the optimal parameter combination; Define logarithmic loss as the loss function and introduce regularization terms, , where TL represents the loss function after the regularization term is introduced. represents the L1 regularization term, Represents the weight of the z-th feature, and takes the number of samples covered when the z-th feature is used as a split point in all trees as the weight of the feature, represents the L2 regularization term, Lo represents the logarithmic loss function; The optimal parameter combination and the initialized model hyperparameters are used as model parameters, and the model parameters are adjusted according to the gradient in each iteration of the model using the training set and loss function; For each iteration, the model parameters are updated using the training set, and the AUC value is calculated on the validation set using AUC as the evaluation metric; A performance improvement threshold xnt is set. If the difference between the AUC value after the iteration and the AUC value of the previous iteration is greater than xnt, the model performance is judged to have improved. If the difference between the AUC value after the iteration and the AUC value of the previous iteration is less than or equal to xnt, the model performance is judged to have not improved. If the model performance on the validation set does not improve in consecutive pat iterations, the training is stopped. The trained gradient boosting decision tree model is obtained.

7. The intelligent packaging production line control system according to claim 6, characterized in that: The method for determining whether a fault exists in the intelligent packaging production line based on the fault probability value of the electrical component includes: Collect historical fault scores, sort them by time, and form a fault score sequence; Set the sliding window HD and the sliding window step hd; Apply the sliding window HD to the fault score sequence. For all fault scores in the initial sliding window, use the ROC curve method to calculate the true positive rate TPR and false positive rate FPR corresponding to each possible fault score threshold. Based on the TPR and FPR within the initial sliding window, the optimal threshold yz is calculated using the Youden index; Move the window forward by a step size nd, recalculate TPR and FPR for all fault scores in the new sliding window, and calculate the optimal threshold yz; Take the average of all the best thresholds in the fault score sequence as the current best threshold; When the fault score is greater than or equal to the optimal threshold, it means that the electrical components of the intelligent packaging production line are faulty. When the fault score is less than the optimal threshold, it means that the electrical components of the intelligent packaging production line are normal. If any electrical component in the intelligent packaging production line fails, it is determined that the intelligent packaging production line has a fault.

8. The intelligent packaging production line control system according to claim 7, characterized in that: When a fault is detected in the intelligent packaging production line, the method of analyzing the faulty electrical components and taking targeted control measures includes: When a fault is detected on the intelligent packaging production line, the faulty electrical component type is analyzed and targeted control measures are taken for different types of electrical components, including: main distribution cabinet, motor inverter, motor driver, central controller and heating element; For the main distribution cabinet, check and adjust the input voltage and current to within the preset normal range, optimize the load distribution within the main distribution cabinet to balance the load on each circuit, and check the setting values ​​of the circuit breakers to ensure that they can respond correctly to abnormal situations; For motor inverters, we can obtain the actual status of the motor and adjust the motor's operating frequency to make the motor run in the best state. We can also optimize the start and stop procedures of the motor inverter to reduce the impact and downtime when the motor starts. For motor drives, adjust the torque output of the motor drive, reduce the load on the motor drive, and optimize the speed setting of the motor to coordinate with other parts of the production line; For the central controller, recalibrate the production parameters in the central controller to make the set production parameters meet the current production requirements; For the heating element, check and adjust the temperature setting value of the heating element to make it within the preset safety range, and calibrate the temperature controller so that the accuracy of temperature measurement is greater than the preset accuracy threshold.

Images