Carrier tape material supply and demand prediction method and system based on machine learning, and storage medium

Abstract

The invention provides a carrier tape material supply and demand prediction method and system based on machine learning and a storage medium, a multi-source perception-intelligent prediction-dynamic control closed loop structure is constructed, equipment physical characteristics, operation specifications and AI prediction are deeply coupled, and unmanned precise management and control of a carrier tape supply scene are realized; the method comprises the following steps: firstly, collecting feeding speed, motor current, temperature and vibration data in real time, and generating a time sequence with a timestamp in combination with a braid type; then, material consumption prediction based on a time sequence neural network and failure rate prediction based on a classifier are carried out; and finally, material supplementing and three-color display are triggered according to the predicted material remaining rate, the operation efficiency of a production line is improved through material prediction, shutdown caused by material breakage is avoided, a fault maintenance mechanism is triggered according to the predicted fault rate in a grading mode, and the maintenance efficiency of faults of hardware such as gears and motors is improved through fault prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of feeder, more particularly, to a tape and reel material supply and demand prediction method, system and storage medium based on machine learning. BACKGROUND

[0002] In the field of precision manufacturing such as semiconductor packaging and SMT mounting, high-speed and stable feeding of tape and reel materials is the core link to ensure the continuity of the production line. The tape and reel materials at least include 32 / 44 / 56 type ribbon. The traditional feeding system relies on manual inspection to judge the remaining amount or uses excessive materials to avoid production line downtime caused by material breakage, which limits the feeding efficiency of the feeder system. For gear wear, motor stall and other faults, there is no obvious sign before they worsen, which leads to sudden downtime and resulting production loss. In addition, in the absence of compensation mechanisms for high temperature, vibration and other environmental factors, the physical quantity collection is biased.

[0003] At present, although sensors are introduced for detection in the prior art, various data such as speed, current, temperature and other parameters are only analyzed independently, and a multi-source correlation model is not established. The static warning based on threshold judgment cannot realize early intervention. Therefore, there is an urgent need for a supply control technology that provides autonomous optimization for the tape and reel feeding scene. SUMMARY

[0004] In view of the above problems, the purpose of the present application is to provide a tape and reel material supply and demand prediction method, system and storage medium based on machine learning, which constructs a closed-loop structure of multi-source perception-intelligent prediction-dynamic control, deeply couples the device physical characteristics, operation specifications and AI prediction, and realizes unmanned and accurate management and control of the tape and reel feeding scene. First, the feeding speed, motor current, temperature and vibration data are collected in real time, and the time-stamped time sequence is generated in combination with the ribbon type. Then, the material consumption prediction based on time series neural network and the fault rate prediction based on classifier are performed. Finally, the material replenishment and three-color display are triggered according to the predicted material remaining rate, the material breakage downtime is avoided through material prediction, the production line operation efficiency is improved, and the fault maintenance mechanism is triggered according to the predicted fault rate classification, the maintenance efficiency of hardware faults such as gears and motors is improved through fault prediction.

[0005] The first aspect of the present application provides a tape and reel material supply and demand prediction method based on machine learning, which comprises: According to a preset collection period, the feeding speed is read based on a preset instruction address, the motor current is collected based on a detection circuit, and the temperature information and vibration information are measured based on a preset environmental sensor; According to the feeding speed and the sampling timestamp, a feeding speed sequence is generated, and a preset rhythm standard deviation information is calculated; Based on a preset time series network model, the material remaining rate is obtained according to the feeding speed sequence and the rhythm standard deviation information. generate a replenishment request instruction or a replenishment warning instruction according to the relationship between the material remaining rate and the remaining rate threshold value; generate a multi-source data sequence according to the motor current, temperature information, vibration information, and timestamp; obtain a fault probability according to the multi-source data sequence based on a preset classifier model; trigger a state display mechanism, a speed adjustment warning mechanism, or an emergency stop mechanism according to the fault probability based on a preset hierarchical early warning mechanism.

[0006] In this scheme, the feeding speed is read based on a preset instruction address according to a preset collection period, the motor current is collected based on a detection circuit, and the temperature information and vibration information are measured based on a preset environment sensor, specifically including: obtain instruction address information of the feeding speed; analyze instruction communication data according to the instruction address information based on a preset communication protocol to obtain the feeding speed; convert to analog-digital information through an analog-digital converter based on a preset current sensor circuit; convert the motor current according to the analog-digital information; measure the real-time temperature value of the feeder to obtain the temperature information according to a preset temperature sensor; measure the vibration amplitude of the feeder motor to obtain the vibration information according to a preset vibration sensor.

[0007] In this scheme, the preset rhythm standard deviation information is calculated, specifically including: cut the feeding speed sequence according to the prediction length of the time series network model; segment the cut feeding speed sequence according to a preset first time window; calculate the speed average and speed variance of the feeding speed sequence in each window; determine the speed tolerance range based on the type of the braid; if the speed average exceeds the speed tolerance range, set first identification information; if the speed variance exceeds a preset variance threshold value, set second identification information; combine the speed average, the speed variance, the first identification information, and the second identification information to form the rhythm standard deviation information.

[0008] In this scheme, the time series network model specifically includes: the input layer receives the feeding speed sequence and the rhythm standard deviation information and performs normalization processing respectively to obtain normalized information; The hidden layer adopts at least 8 layers of time convolution to extract local feature information; The full connection layer sets a type code vector according to a type of the tape; The output layer obtains the material remaining rate based on a preset mapping network according to the normalized information, the local feature information and the type code vector.

[0009] In the scheme, the material remaining rate and the remaining rate threshold value are related to generate a replenishment request instruction or a replenishment warning instruction, specifically: The remaining rate threshold value at least includes a first remaining threshold value and a second remaining threshold value; If the material remaining rate is greater than the first remaining threshold value, a green progress bar is displayed through a man-machine interaction interface; If the material remaining rate is in the interval range of the first remaining threshold value and the second remaining threshold value, a replenishment request instruction is generated, and a yellow and flashing progress bar is displayed through the man-machine interaction interface according to the replenishment request instruction; If the material remaining rate is less than the second remaining threshold value, a replenishment warning instruction is generated, and a red progress bar is displayed through the man-machine interaction interface according to the replenishment warning instruction and a replenishment confirmation pop-up window is popped up.

[0010] In the scheme, based on a preset hierarchical early warning mechanism, the state display mechanism, the speed adjustment warning mechanism or the emergency stop mechanism is triggered according to the fault probability, specifically: If the fault probability is less than a preset first fault threshold value, the state display mechanism is triggered, and a normal state icon is displayed through a man-machine interaction interface; If the fault probability is in the interval range of the preset first fault threshold value and a preset second fault threshold value, the speed adjustment warning mechanism is triggered, a fault warning state icon is displayed through the man-machine interaction interface, and the feeding speed is adjusted to a preset safe speed or switched to a standby step distance; If the fault probability is greater than the preset second fault threshold value, the emergency stop mechanism is triggered, a maintenance work order is displayed through the man-machine interaction interface, and the driving power supply of the feeding motor is disconnected.

[0011] The second aspect of the present application provides a tape material supply and demand prediction system based on machine learning, which comprises a tape material supply and demand prediction method program based on machine learning. According to a preset collection period, the feeding speed is read based on a preset instruction address, the motor current is collected based on a detection circuit, and the temperature information and vibration information are measured based on a preset environment sensor; According to the feeding speed and the sampling time stamp, a feeding speed sequence is generated, and preset rhythm standard deviation information is calculated. Based on the preset timing network model, the material remaining rate is obtained according to the feeding speed sequence and the rhythm standard deviation information; According to the relationship between the material remaining rate and the remaining rate threshold value, a replenishment request instruction or a replenishment warning instruction is generated; According to the motor current, temperature information, vibration information and time stamp, a multi-source data sequence is generated; Based on the preset classifier model, the fault probability is obtained according to the multi-source data sequence; Based on the preset hierarchical early warning mechanism, the state display mechanism, the speed adjustment warning mechanism or the emergency stop mechanism is triggered according to the fault probability.

[0012] In the scheme, the feeding speed is read based on the preset instruction address according to the preset collection period, the motor current is collected based on the detection circuit, and the temperature information and the vibration information are measured based on the preset environment sensor, which specifically includes: The instruction address information of the feeding speed is obtained; Based on the preset communication protocol, the instruction communication data is parsed according to the instruction address information to obtain the feeding speed; Based on the preset current sensor circuit, analog-digital conversion is performed by an analog-digital converter to obtain analog-digital information; The motor current is converted according to the analog-digital information; According to the preset temperature sensor, the real-time temperature value of the feeder is measured to obtain the temperature information; According to the preset vibration sensor, the vibration amplitude of the feeder motor is measured to obtain the vibration information.

[0013] In the scheme, the preset rhythm standard deviation information is calculated, which specifically includes: According to the prediction length of the timing network model, the feeding speed sequence is intercepted; According to the preset first time window, the intercepted feeding speed sequence is segmented; The speed average and speed variance of the feeding speed sequence in each window are calculated; The speed tolerance range is determined based on the braid type; If the speed average exceeds the speed tolerance range, the first identification information is set; If the speed variance exceeds the preset variance threshold value, the second identification information is set; The speed average, the speed variance, the first identification information and the second identification information are combined to form the rhythm standard deviation information.

[0014] The third aspect of the present application provides a computer readable storage medium, wherein the computer readable storage medium comprises a machine learning-based tape material supply and demand prediction method program, and the machine learning-based tape material supply and demand prediction method program is executed by a processor to implement the steps of the machine learning-based tape material supply and demand prediction method according to any one of the above aspects.

[0015] The present application provides a machine learning-based tape material supply and demand prediction method, system and storage medium, constructs a multi-source perception-intelligent prediction-dynamic control closed loop structure, deeply couples device physical characteristics, operation specifications and AI prediction, and realizes unmanned precise management and control of the tape supply scene; first, real-time collection of feeding speed, motor current, temperature and vibration data, combination with the ribbon type to generate a time-stamped time sequence; then, material consumption prediction based on a time sequence neural network and fault rate prediction based on a classifier; finally, triggering of material replenishment and three-color display according to the predicted material remaining rate, avoiding material breakage downtime through material prediction, improving production line operation efficiency, and triggering of a fault maintenance mechanism according to the predicted fault rate classification, improving the maintenance efficiency of hardware faults such as gears and motors through fault prediction. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope.

[0017] Figure 1 A flowchart of a machine learning-based tape material supply and demand prediction method of the present application is shown; Figure 2 A multi-source data acquisition flowchart provided by an embodiment of the present application is shown; Figure 3 A rhythm standard deviation information generation flowchart provided by an embodiment of the present application is shown; Figure 4 A block diagram of a machine learning-based tape material supply and demand prediction system of the present application is shown. DETAILED DESCRIPTION

[0018] The technical solutions of the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0019] Unless otherwise defined, all terms (including technical and scientific terms) used in the present application embodiments have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined in the present application embodiments.

[0020] The terms "first", "second", and similar terms used in the present application embodiments do not denote any order, quantity, or importance, but are used to distinguish different constituent elements. The terms "one", "a", or "the" and similar terms do not denote a quantity of particular mentioned items, but mean that there is at least one of the items present. The terms "comprises", "comprising", or "including" and similar terms mean that the elements listed after the word encompass the elements recited therein and equivalents thereof, and do not exclude other elements. The terms "connected", "coupled", or "linked" and similar terms are not limited to physical or mechanical connections or couplings, but can also include electrical connections or couplings, whether direct or indirect. The steps in the methods of the present application embodiments need not be performed in the order indicated in the present application embodiments. Conversely, various steps can be handled in reverse order, or at the same time. Other operations can also be added to the processes, or steps can be removed from the processes.

[0021] In addition, the functional modules in each of the embodiments of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0022] Figure 1 A flowchart of a machine learning-based carrier tape material supply and demand prediction method of the present application is shown.

[0023] As shown in Figure 1 The first aspect of the present application discloses a machine learning-based carrier tape material supply and demand prediction method, which comprises: S102, reading the feeding speed based on a preset instruction address, collecting the motor current based on a detection circuit, and measuring temperature information and vibration information based on a preset environment sensor according to a preset collection period; S104, generating a feeding speed sequence according to the feeding speed and a sampling timestamp, and calculating preset rhythm standard deviation information; S106, obtaining a material remaining rate based on a preset timing network model according to the feeding speed sequence and the rhythm standard deviation information; S108, generating a replenishment request instruction or a replenishment warning instruction according to the relationship between the material remaining rate and a remaining rate threshold; S110, generating a multi-source data sequence according to the motor current, temperature information, vibration information and time stamp; S112, obtaining a fault probability according to the multi-source data sequence based on a preset classifier model; S114, triggering a state display mechanism, a speed adjustment warning mechanism or an emergency stop mechanism according to the fault probability based on a preset hierarchical warning mechanism.

[0024] It should be noted that the embodiment provides a tape material supply and demand prediction process based on machine learning. The rhythm standard deviation information is used to reflect the structured data of the feeding state. In this embodiment, multi-source data is collected at a preset data collection period, for example, at a period of 50 ms. As an implementation, the feeder is provided with a 12Pin aviation connector for connecting data communication lines and sensor data lines. The feeding speed is read in real time through the data communication line according to the communication address, the motor driving current is collected in real time by using an analog-to-digital converter and a current collection circuit, and the temperature value of the feeder and the vibration condition of the feeder motor are detected in real time by using a temperature sensor and a vibration sensor. First, a feeding speed sequence is generated based on the feeding speed data according to the time stamp, and rhythm standard deviation information is obtained according to a preset time sequence feature marking mechanism. The feeding speed sequence and the rhythm standard deviation information are input into a preset time sequence network model to predict the material remaining rate. According to the relationship between the predicted material remaining rate and the preset remaining rate threshold, corresponding replenishment or warning operations are performed; as an implementation, when the material remaining rate is too low, a replenishment warning instruction is issued to avoid material breakage; when the material remaining rate is within the allowable replenishment range, a replenishment request instruction is generated to reduce the replenishment time. Then, the motor current, temperature information, vibration information and the like are grouped into a multi-source data sequence according to the time stamp. The multi-source data sequence is input into a preset classifier model to predict the fault probability of the feeder. According to the fault probability, corresponding fault warning operations are performed based on a preset fault hierarchical warning mechanism; as an implementation, when the fault rate is too high, alarm and emergency stop are used to avoid greater interruption loss caused by continuous operation of the accessory; when the fault rate is within the allowable warning range, the feeding speed is adjusted to avoid aggravating the mechanical failure of the feeder accessory. The embodiment combines device physical parameters and environmental parameters to predict the material remaining rate and the fault probability through multi-source data combination, so as to reduce the influence of line material breakage or mechanical failure on production efficiency.

[0025] Figure 2 A multi-source data collection flowchart provided by the embodiment of the application is shown.

[0026] According to the embodiment of the application, Figure 2As shown, the preset acquisition cycle is based on the preset instruction address reading speed, based on the detection circuit acquisition motor current, based on the preset environmental sensor measurement temperature information and vibration information, specifically comprising: S202, obtaining instruction address information of the feeding speed; S204, based on the preset communication protocol, according to the instruction address information, analyzing instruction communication data, obtaining the feeding speed; S206, based on the preset current sensor circuit, converting to obtain analog-digital information through an analog-digital converter; S208, according to the analog-digital information, converting to obtain the motor current; S210, according to the preset temperature sensor, measuring the real-time temperature value of the feeder, obtaining the temperature information; S212, according to the preset vibration sensor, measuring the vibration amplitude of the feeder motor, obtaining the vibration information.

[0027] It should be noted that, as an embodiment, through the preset RS232 protocol, the data communication is carried out according to the instruction address of the feeding speed, the speed value in the returned data frame is analyzed, and then the feeding speed is obtained. In addition, the GPIO connecting the two parties in the 12Pin aviation connector is set as the communication control pin to prevent data misreading caused by interference signals. For current signal acquisition, the current sensor is connected in series to the motor power supply circuit, and the analog-digital conversion value (AD value) is obtained through 24-bit high-precision analog-digital conversion (ADC) sampling conversion, and then the motor current is obtained according to the AD-current relationship of the current sensor. Through the temperature sensor and the vibration sensor, the temperature information and the vibration information output by the sensor are obtained based on the preset communication mode or analog quantity detection mode.

[0028] Figure 3 A rhythm standard deviation information generation flowchart provided by an embodiment of the application is shown.

[0029] According to the embodiment of the application, as Figure 3 shown, the preset rhythm standard deviation information is calculated, specifically comprising: S302, according to the prediction length of the time sequence network model, intercepting the feeding speed sequence; S304, according to the preset first time window, segmenting the intercepted feeding speed sequence; S306, calculating the speed average and speed variance of the feeding speed sequence in each window; S308, determining the speed tolerance range based on the braiding type; S310, if the speed average exceeds the speed tolerance range, setting the first identification information; S312, if the speed variance exceeds a preset variance threshold, setting second identification information; S314, combining the speed average, the speed variance, the first identification information and the second identification information, composing the rhythm standard deviation information.

[0030] It should be noted that the embodiment provides a segmentation processing mechanism of a time sequence. As an implementation manner, 30 minutes are taken as a prediction time length of a time sequence network model, a feeding speed sequence in the last 30 minutes is intercepted, 15 seconds are taken as a first time window, and the intercepted feeding speed sequence is segmented. The average value and the variance of the feeding speed in each window are calculated respectively. Then, based on the type of the conveying material, a conveying speed tolerance range allowed by the ribbon is determined, when the average value of the feeding speed exceeds the conveying speed tolerance range, the first identification information is set for the corresponding time window, and the overspeed label is indicated. According to the relationship between the speed variance and the variance threshold, the second identification information is set, wherein when the speed variance exceeds the variance threshold, it indicates that the speed fluctuation range is large, and the second identification information is set to indicate that the feeding speed has abnormal fluctuation. The speed average, the speed variance, the first identification information and the second identification information of each time window are combined to compose the rhythm standard deviation information, so as to reflect the time sequence characteristics of the feeding speed.

[0031] According to the embodiment of the present application, the time sequence network model specifically comprises: The input layer receives the feeding speed sequence and the rhythm standard deviation information, and performs normalization processing respectively to obtain normalized information; The hidden layer adopts at least 8 layers of time sequence convolution to extract local feature information; The full connection layer sets a type code vector according to the type of the ribbon; The output layer obtains the material residual rate based on a preset mapping network according to the normalized information, the local feature information and the type code vector.

[0032] It should be noted that the embodiment provides a running mechanism of the time sequence network model. As an implementation, an LSTM time sequence model is adopted, and the time sequence network model includes an input layer, a hidden layer, a full connection layer, and an output layer. The input layer adopts a Sigmoid function to normalize the time sequence characteristics reflected by the feeding speed and the rhythm standard deviation information, and the influence degree of each time sequence characteristic is reflected through the [0, 1] interval. The hidden layer processes the feeding speed sequence and the rhythm standard deviation information through at least 8 layers of time sequence convolution, so as to extract a local feature vector. The full connection layer sets a type coding vector according to the type of the braid, for example, the type coding vector of the 32 type braid is [1, 0, 0], the type coding vector of the 44 type braid is [0, 1, 0], and the type coding vector of the 56 type braid is [0, 0, 1]. The output layer calculates the material remaining rate based on the pre-trained weight parameters according to the normalized data, the local feature vector, and the type coding vector.

[0033] According to the embodiment of the present application, the replenishment request instruction or the replenishment warning instruction is generated according to the relationship between the material remaining rate and the remaining rate threshold value, specifically: The remaining rate threshold value at least includes a first remaining threshold value and a second remaining threshold value; If the material remaining rate is greater than the first remaining threshold value, a green progress bar is displayed through the man-machine interaction interface; If the material remaining rate is within the interval range of the first remaining threshold value and the second remaining threshold value, a replenishment request instruction is generated, and a yellow and flashing progress bar is displayed through the man-machine interaction interface according to the replenishment request instruction; If the material remaining rate is less than the second remaining threshold value, a replenishment warning instruction is generated, and a red progress bar is displayed through the man-machine interaction interface according to the replenishment warning instruction, and a replenishment confirmation pop-up window is popped up.

[0034] It should be noted that the embodiment provides a three-color display mechanism based on the material remaining rate. As an implementation, the first remaining threshold value is 30%, the second remaining threshold value is 10%, and the touch display screen is used as the man-machine interaction interface. When the predicted material remaining rate is greater than 30%, it indicates that the material is sufficient, and a green progress bar is displayed in the material remaining column through the touch display screen, to indicate that the material is sufficient. When the predicted material remaining rate is within the interval [10%, 30%], it indicates that the material remaining amount is low, and a yellow flashing progress bar is displayed in the material remaining column through the touch display screen, to prompt attention to replenishment. When the predicted material remaining rate is less than 10%, it indicates that the material remaining amount is too low and is at a risk level, a replenishment warning instruction is generated, a red progress bar is displayed in the material remaining column through the touch display screen, and a replenishment confirmation pop-up window is displayed, to alert the replenishment demand. The three-color display reduces the operator's inspection frequency, reduces the probability of missing detection, and further reduces the production of finished products.

[0035] According to the embodiment of the present application, the preset hierarchical early warning mechanism is used to trigger a state display mechanism, a speed adjustment warning mechanism or an emergency stop mechanism according to the failure probability, in particular: If the failure probability is less than a preset first failure threshold, the state display mechanism is triggered to display a normal state icon through a human-computer interaction interface; If the failure probability is within a range of the preset first failure threshold and a preset second failure threshold, the speed adjustment warning mechanism is triggered to display a failure warning state icon through the human-computer interaction interface, and the feeding speed is adjusted to a preset safe speed or switched to a backup step distance; If the failure probability is greater than the preset second failure threshold, the emergency stop mechanism is triggered to display a maintenance work order through the human-computer interaction interface, and the driving power supply of the feeding motor is disconnected.

[0036] It should be noted that the embodiment provides a failure hierarchical early warning mechanism. As an implementation manner, the first failure threshold is 60% and the second failure threshold is 80%. When the predicted failure probability is less than 60%, a green gear icon is displayed through the display screen, indicating that the failure probability is low. When the predicted failure probability is within the range of [60, 80%], an orange gear icon is displayed, indicating that there is a failure risk, and speed reduction protection is performed based on the preset safe running speed or switched to a backup coding step distance to reduce the mechanical load of the motor, thereby avoiding aggravating the mechanical failure of the feeder accessory. When the predicted failure probability is greater than 80%, a red gear icon is displayed and a maintenance work order is popped up, indicating that the failure risk is large and maintenance needs to be performed, and the motor driving power supply is cut off through the emergency stop relay and other modules to reduce the mechanical loss of the motor continuous operation.

[0037] It is worth mentioning that the preset classifier model is used to obtain the failure probability according to the multi-source data sequence, in particular: The preset temperature compensation mapping relationship is used to obtain a compensation current according to the temperature information and the motor current; The preset current integration period is used to calculate a current change rate per unit time according to the compensation current; The current change rate is used to identify a current sudden change abnormal point; The vibration information is used to calculate an energy proportion in a preset frequency band; The energy proportion is used to identify a vibration abnormal point; The multi-source data sequence, the current sudden change abnormal point and the vibration abnormal point are used to match a prediction mechanism in the classifier to obtain the failure probability.

[0038] It should be noted that the present application provides a classifier model prediction mechanism. First, according to the relationship between the motor current and the temperature value, the current compensation value is calculated based on the current temperature value to obtain the compensated motor current value, which is recorded as the compensation current. Then, the integral value of the compensation current is calculated based on the preset unit time, and the change in the current integral value per unit time is calculated to obtain the current change rate. When the absolute value of the current change rate exceeds the preset change rate threshold, it indicates that there is a mutation in the current, and it is determined to be a mutation anomaly point. Then, a 512-point FFT transform is performed on the vibration signal to calculate the energy proportion in the frequency band from 0.5KHz to 2KHz. The vibration anomaly point is identified and judged based on the energy proportion. Combined with multi-source data sequences, current mutation anomaly points and vibration anomaly points, they are input into a pre-trained classifier, such as a random forest model, an XGBoost classifier, etc., to obtain the fault probability.

[0039] It is worth mentioning that it also includes a self-learning update mechanism, specifically: Calculate the error between the predicted material surplus rate and the actual surplus rate based on the preset self-learning update cycle; If the error is greater than a preset first deviation threshold, the training data set is injected after setting the weighting coefficient; Extract multi-source data sequences of a preset length before the fault event, set labels, and use them to expand the classifier samples.

[0040] It should be noted that the error between the predicted surplus rate and the actual surplus rate is statistically calculated in real time, with each day being the time unit or 1,000 groups of cumulative feeding being the self-learning update cycle. If the error exceeds the preset first deviation threshold, the weight is set based on the error. As an implementation method, when the error value is greater than 5%, it is marked as a weighted sample. The weighting coefficient and the feeding speed sequence corresponding to the sample are used as a training data set to train the timing network model. In addition, when a fault occurs, the multi-source data sequence 30 minutes before the fault time is extracted, marked with the corresponding fault label, such as gear fault, current overload, etc., and then added to the classifier sample library. This embodiment uses an error weighting mechanism to enable the model to adapt to production line changes and improve the accuracy of the prediction.

[0041] Figure 4 A block diagram of a carrier material supply and demand forecasting system based on machine learning is shown in the present invention.

[0042] like Figure 4 As shown, the second aspect of the present invention discloses a carrier material supply and demand forecasting system 4 based on machine learning, including a memory 41 and a processor 42. The memory includes a carrier material supply and demand forecasting method program based on machine learning. When the carrier material supply and demand forecasting method program based on machine learning is executed by the processor, the following steps are implemented: According to a preset acquisition cycle, reading the feeding speed based on a preset instruction address, collecting the motor current based on a detection circuit, measuring temperature information and vibration information based on a preset environment sensor; According to the feeding speed and the sampling timestamp, generating a feeding speed sequence, and calculating preset rhythm standard deviation information; Based on a preset timing network model, according to the feeding speed sequence and the rhythm standard deviation information, obtaining the material remaining rate; According to the relationship between the material remaining rate and the remaining rate threshold, generating a replenishment request instruction or a replenishment warning instruction; According to the motor current, temperature information, vibration information and timestamp, a multi-source data sequence is generated; Based on a preset classifier model, according to the multi-source data sequence, the failure probability is obtained; Based on a preset hierarchical early warning mechanism, according to the failure probability, triggering a state display mechanism, a speed adjustment warning mechanism or an emergency stop mechanism.

[0043] It should be noted that the embodiment provides a tape material supply and demand prediction process based on machine learning. The rhythm standard deviation information is used to reflect the structured data of the feeding state. In this embodiment, multi-source data is collected at a preset data collection period, for example, at a period of 50 ms. As an implementation, a 12Pin aviation connector is provided for the feeder to access data communication lines and sensor data lines. The feeding speed is read in real time through the data communication line according to the communication address, the motor driving current is collected in real time by using an analog-to-digital converter and a current collection circuit, and the temperature value of the feeder and the vibration condition of the feeder motor are detected in real time by using a temperature sensor and a vibration sensor. First, a feeding speed sequence is generated based on the feeding speed data according to the time stamp, and the rhythm standard deviation information is obtained according to a preset timing feature marking mechanism. The feeding speed sequence and the rhythm standard deviation information are input into a preset timing network model to predict the material remaining rate. According to the relationship between the predicted material remaining rate and the preset remaining rate threshold, a corresponding replenishment or warning operation is performed; as an implementation, when the material remaining rate is too low, a replenishment warning instruction is issued to avoid material breakage; when the material remaining rate is within the allowable replenishment range, a replenishment request instruction is generated to reduce the replenishment time. Then, the motor current, temperature information, vibration information and other data are grouped into a multi-source data sequence according to the time. The multi-source data sequence is input into a preset classifier model to predict the fault probability of the feeder. According to the fault probability, a corresponding fault warning operation is performed based on a preset fault grading warning mechanism; as an implementation, when the fault rate is too high, alarm and emergency stop are adopted to avoid greater interruption loss caused by continuous operation of the accessory; when the fault rate is within the allowable warning range, the feeding speed is adjusted to avoid aggravating the mechanical failure of the feeder accessory. In this embodiment, the physical parameters and environmental parameters of the equipment are fused, and the multi-source data is combined to predict the material remaining rate and the fault probability, so as to reduce the influence of line material breakage or mechanical failure on production efficiency.

[0044] According to the embodiment of the present application, the feeding speed is read based on a preset instruction address according to a preset collection period, the motor current is collected based on a detection circuit, and the temperature information and vibration information are measured based on a preset environmental sensor. Specifically, the method comprises the following steps: obtaining instruction address information of the feeding speed; based on a preset communication protocol, analyzing instruction communication data according to the instruction address information to obtain the feeding speed; based on a preset current sensor circuit, converting the motor current by using an analog-to-digital converter to obtain analog information; converting the motor current according to the analog information; measuring the real-time temperature value of the feeder according to a preset temperature sensor to obtain the temperature information; According to the preset vibration sensor, the vibration amplitude of the feeder motor is measured to obtain the vibration information.

[0045] It should be noted that, as an embodiment, the RS232 protocol is preset, the data communication is performed according to the instruction address of the feeding speed, the speed value in the returned data frame is analyzed, and then the feeding speed is obtained. In addition, the GPIO connecting the two communication parties in the 12Pin aviation connector is set as a communication control pin to prevent data misreading caused by interference signals. For current signal acquisition, the current sensor is connected in series to the motor power supply circuit, and the analog-to-digital conversion value (AD value) is obtained through 24-bit high-precision analog-to-digital conversion (ADC). Then, according to the AD-current relationship of the current sensor, the motor current is obtained. Through the temperature sensor and the vibration sensor, the temperature information and the vibration information output by the sensor are obtained based on the preset communication mode or analog quantity detection mode.

[0046] According to the embodiment of the present application, the calculated preset rhythm standard deviation information specifically includes: According to the predicted time length of the time sequence network model, the feeding speed sequence is intercepted; According to the preset first time window, the intercepted feeding speed sequence is segmented; The speed average and the speed variance of the feeding speed sequence in each window are calculated; Based on the type of the braid, the speed tolerance range is determined; If the speed average exceeds the speed tolerance range, the first identification information is set; If the speed variance exceeds the preset variance threshold, the second identification information is set; The speed average, the speed variance, the first identification information and the second identification information are combined to form the rhythm standard deviation information.

[0047] It should be noted that the embodiment provides a segmentation processing mechanism of a time sequence. As an implementation, the feeding speed sequence of the last 30 minutes is intercepted with 30 minutes as the prediction time length of the time sequence network model, and the intercepted feeding speed sequence is segmented with 15 seconds as a first time window. The average value and the variance of the feeding speed in each window are calculated respectively. Then, based on the type of the tape for conveying the material, the allowable conveying speed range of the tape is determined, and when the average value of the feeding speed exceeds the allowable conveying speed range, the corresponding time window is set with first identification information, which indicates an overspeed label. According to the relationship between the speed variance and the variance threshold, second identification information is set, wherein when the speed variance exceeds the variance threshold, it indicates that the speed fluctuation range is large, and the second identification information is set to indicate that the feeding speed has abnormal fluctuation. The speed average value, the speed variance, the first identification information and the second identification information of each time window are combined to form rhythm standard deviation information, which is used to reflect the time sequence characteristics of the feeding speed.

[0048] According to the embodiment of the present application, the time sequence network model specifically comprises: The input layer receives the feeding speed sequence and the rhythm standard deviation information, and performs normalization processing respectively to obtain normalized information; The hidden layer adopts at least 8 layers of time sequence convolution to extract local feature information; The full connection layer sets a type code vector according to the type of the tape; The output layer obtains the material remaining rate based on a preset mapping network according to the normalized information, the local feature information and the type code vector.

[0049] It should be noted that the embodiment provides a running mechanism of the time sequence network model. As an implementation, an LSTM time sequence model is adopted, and the time sequence network model comprises an input layer, a hidden layer, a full connection layer and an output layer. The input layer adopts a Sigmoid function to perform normalization processing on the time sequence characteristics reflected by the feeding speed and the rhythm standard deviation information, and reflects the influence degree of each time sequence characteristic through the [0, 1] interval. The hidden layer processes the feeding speed sequence and the rhythm standard deviation information through at least 8 layers of time sequence convolution, so as to extract a local feature vector. The full connection layer sets a type code vector according to the type of the tape, for example, the type code vector of the 32 type tape is [1, 0, 0], the type code vector of the 44 type tape is [0, 1, 0], and the type code vector of the 56 type tape is [0, 0, 1]. The output layer calculates the material remaining rate based on the pre-trained weight parameters according to the normalized data, the local feature vector and the type code vector.

[0050] According to the embodiment of the present application, the material replenishment request instruction or the material replenishment warning instruction is generated according to the relationship between the material remaining rate and the remaining rate threshold, specifically: The residual rate threshold value comprises at least a first residual threshold value and a second residual threshold value; If the material residual rate is greater than the first residual threshold value, a green progress bar is displayed through the man-machine interactive interface; If the material residual rate is within the interval range of the first residual threshold value and the second residual threshold value, a replenishment request instruction is generated, and a yellow and flashing progress bar is displayed through the man-machine interactive interface according to the replenishment request instruction; If the material residual rate is less than the second residual threshold value, a replenishment warning instruction is generated, and a red progress bar is displayed through the man-machine interactive interface according to the replenishment warning instruction, and a replenishment confirmation pop-up window is popped up.

[0051] It should be noted that the embodiment provides a three-color display mechanism based on the material residual rate. As an implementation manner, the first residual threshold value is 30%, the second residual threshold value is 10%, and the touch display screen is used as the man-machine interactive interface. When the predicted material residual rate is greater than 30%, it indicates that the material is sufficient, and a green progress bar is displayed in the material residual column through the touch display screen, which is used to indicate that the material is sufficient. When the predicted material residual rate is within the interval [10%, 30%], it indicates that the material residual amount is low, and a yellow flashing progress bar is displayed in the material residual column through the touch display screen, which is used to prompt attention to replenishment. When the predicted material residual rate is less than 10%, it indicates that the material residual amount is too low and is at a risk level, and a replenishment warning instruction is generated, a red progress bar is displayed in the material residual column through the touch display screen, and a replenishment confirmation pop-up window is displayed, which is used to alert the replenishment demand. Through the three-color display, the operator's inspection frequency is reduced, the missed detection probability is reduced, and the production finished product is reduced.

[0052] According to the embodiment of the present application, based on the preset hierarchical early warning mechanism, according to the fault probability, a state display mechanism, a speed adjustment warning mechanism or an emergency stop mechanism is triggered, specifically: If the fault probability is less than a preset first fault threshold value, the state display mechanism is triggered, and a normal state icon is displayed through the man-machine interactive interface; If the fault probability is within the interval range of the preset first fault threshold value and the preset second fault threshold value, the speed adjustment warning mechanism is triggered, a fault warning state icon is displayed through the man-machine interactive interface, and the feeding speed is adjusted to a preset safe speed or switched to a standby step distance; If the fault probability is greater than the preset second fault threshold value, the emergency stop mechanism is triggered, a maintenance work order is displayed through the man-machine interactive interface, and the driving power supply of the feeding motor is disconnected.

[0053] It should be noted that this embodiment provides a fault classification warning mechanism. As an implementation method, the first fault threshold is 60% and the second fault threshold is 80%. When the predicted failure probability is lower than 60%, a green gear icon is displayed on the display screen, indicating that the failure probability is low. When the predicted failure probability is within the range of [60,80%], an orange gear icon is displayed, indicating that there is a failure risk, and speed reduction protection is performed based on the preset safe operating speed or switching to the spare taping step is performed to achieve the purpose of reducing the motor load, thereby avoiding aggravating the mechanical failure of the feeder accessories. When the predicted failure probability is greater than 80%, a red gear icon is displayed and a maintenance work order pops up, indicating that the failure risk is high and maintenance needs to be performed, and the motor drive power is cut off through modules such as the emergency stop relay to reduce the mechanical loss of the motor during continuous operation.

[0054] It is worth mentioning that the fault probability is obtained based on the preset classifier model according to the multi-source data sequence, specifically: Based on a preset temperature compensation mapping relationship, a compensation current is obtained according to the temperature information and the motor current; Based on a preset current integration period, the current change rate per unit time is calculated according to the compensation current; Identifying abnormal current mutation points based on the current change rate; Calculating the energy ratio within a preset frequency band based on the vibration information; Identify vibration abnormal points based on the energy proportion; According to the multi-source data sequence, the current mutation abnormal point and the vibration abnormal point, a prediction mechanism in a classifier is matched to obtain the fault probability.

[0055] It should be noted that the present application provides a classifier model prediction mechanism. First, according to the relationship between the motor current and the temperature value, the current compensation value is calculated based on the current temperature value to obtain the compensated motor current value, which is recorded as the compensation current. Then, the integral value of the compensation current is calculated based on the preset unit time, and the change in the current integral value per unit time is calculated to obtain the current change rate. When the absolute value of the current change rate exceeds the preset change rate threshold, it indicates that there is a mutation in the current, and it is determined to be a mutation anomaly point. Then, a 512-point FFT transform is performed on the vibration signal to calculate the energy proportion in the frequency band from 0.5KHz to 2KHz. The vibration anomaly point is identified and judged based on the energy proportion. Combined with multi-source data sequences, current mutation anomaly points and vibration anomaly points, they are input into a pre-trained classifier, such as a random forest model, an XGBoost classifier, etc., to obtain the fault probability.

[0056] It is worth mentioning that it also includes a self-learning update mechanism, specifically: According to a preset self-learning update period, an error between a predicted residual rate and an actual residual rate is calculated; If the error is greater than a preset first deviation threshold, a training data set is injected after setting a weighting coefficient; A multi-source data sequence in a preset time period before a fault event is extracted, and a label is set to expand the classifier sample.

[0057] It should be noted that the error between the predicted residual rate and the actual residual rate is calculated in real time with a daily time unit or with 1000 groups of cumulative feeding as the self-learning update period. If the error exceeds the preset first deviation threshold, a weight is set based on the error. As an implementation, when the error value is greater than 5%, the weight sample is marked. The weighting coefficient and the corresponding feeding speed sequence of the sample are used as the training data set to train the time series network model. In addition, when a fault occurs, the multi-source data sequence in the 30 minutes before the fault time is extracted, and the corresponding fault label such as gear failure, current overload, etc. is marked, and then added to the classifier sample library. The error weighting mechanism in this embodiment makes the model adapt to the production line change and improves the prediction accuracy.

[0058] The third aspect of the present application provides a computer readable storage medium, the computer readable storage medium comprises a machine learning based tape material supply and demand prediction method program, when the machine learning based tape material supply and demand prediction method program is executed by a processor, the steps of the machine learning based tape material supply and demand prediction method are realized.

[0059] The present application provides a machine learning based tape material supply and demand prediction method, system and storage medium, which constructs a multi-source perception-intelligent prediction-dynamic control closed loop structure, deeply couples the device physical characteristics, operation specifications and AI prediction, realizes unmanned precise control of the tape supply scene; first, the feeding speed, motor current, temperature, vibration data are collected in real time, and the time-stamped time series sequence is generated combined with the ribbon type; then, the material consumption prediction based on the time series neural network and the fault rate prediction based on the classifier; finally, according to the predicted material residual rate, the feeding and three-color display are triggered, through material prediction, the material breakage downtime is avoided, and the production line operation efficiency is improved, and according to the predicted fault rate grading, the fault maintenance mechanism is triggered, through fault prediction, the maintenance efficiency of gear, motor and other hardware faults is improved.

[0060] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the parts of the technical solutions that essentially contribute to the prior art or the parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0061] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting supply and demand of carrier materials based on machine learning, characterized in that: The method comprises: According to the preset acquisition cycle, the feeding speed is read based on the preset instruction address, the motor current is collected based on the detection circuit, and the temperature information and vibration information are measured based on the preset environmental sensor; Generate a feeding speed sequence according to the feeding speed and sampling timestamp, and calculate and obtain preset rhythm standard deviation information; Based on a preset timing network model, the material remaining rate is obtained according to the feeding speed sequence and the rhythm standard deviation information; generating a material replenishment request instruction or a material replenishment warning instruction according to the relationship between the material remaining rate and the remaining rate threshold; generating a multi-source data sequence according to the motor current, temperature information, vibration information and timestamp; Obtaining a fault probability based on the multi-source data sequence based on a preset classifier model; Based on the preset hierarchical warning mechanism, the status display mechanism, speed regulation warning mechanism or emergency stop mechanism is triggered according to the failure probability.

2. The method for predicting supply and demand of carrier materials based on machine learning according to claim 1, characterized in that: The method includes: reading the feeding speed based on a preset instruction address according to a preset acquisition cycle, collecting the motor current based on a detection circuit, and measuring the temperature information and vibration information based on a preset environmental sensor. Get the instruction address information of feeding speed; Based on a preset communication protocol and according to the instruction address information, the instruction communication data is parsed to obtain the feeding speed; Based on the preset current sensor circuit, analog-to-digital information is converted by an analog-to-digital converter; Converting the motor current according to the analog-to-digital information; According to the preset temperature sensor, the real-time temperature value of the feeder is measured to obtain the temperature information; The vibration amplitude of the feeder motor is measured according to a preset vibration sensor to obtain the vibration information.

3. The method for predicting supply and demand of carrier materials based on machine learning according to claim 1, characterized in that: The calculation to obtain the preset rhythm standard deviation information specifically includes: According to the predicted duration of the timing network model, the feeding speed sequence is intercepted; Segmenting the intercepted feeding speed sequence according to a preset first time window; Calculate the speed average and speed variance of the feeding speed sequence in each window; Determine the speed tolerance based on the tape type; If the speed average value exceeds the speed allowable range, setting first identification information; If the speed variance exceeds a preset variance threshold, setting second identification information; The tempo standard deviation information is composed by combining the speed average, the speed variance, the first identification information, and the second identification information.

4. The method for predicting supply and demand of carrier materials based on machine learning according to claim 1, characterized in that: The temporal network model specifically includes: The input layer receives the feeding speed sequence and the rhythm standard deviation information, and performs normalization processing on them respectively to obtain normalized information; The hidden layer uses at least 8 layers of temporal convolution to extract local feature information; The fully connected layer sets the type encoding vector according to the braiding type; The output layer obtains the material surplus rate according to the normalized information, the local feature information and the type coding vector based on a preset mapping network.

5. The method for predicting supply and demand of carrier materials based on machine learning according to claim 1, characterized in that: The generating of a material replenishment request instruction or a material replenishment warning instruction according to the relationship between the material remaining rate and the remaining rate threshold is specifically as follows: The remaining rate threshold includes at least a first remaining threshold and a second remaining threshold; If the material remaining rate is greater than the first remaining threshold, a green progress bar is displayed through the human-computer interaction interface; If the material remaining rate is within the interval between the first remaining threshold and the second remaining threshold, a material replenishment request instruction is generated, and a yellow and flashing progress bar is displayed through the human-computer interaction interface according to the material replenishment request instruction; If the material remaining rate is less than the second remaining threshold, a refill warning instruction is generated, and a red progress bar is displayed through the human-computer interaction interface according to the refill warning instruction and a refill confirmation pop-up window is popped up.

6. The method for predicting supply and demand of carrier materials based on machine learning according to claim 1, characterized in that: The preset hierarchical warning mechanism triggers the status display mechanism, speed regulation warning mechanism or emergency stop mechanism according to the failure probability, specifically: If the failure probability is less than a preset first failure threshold, a status display mechanism is triggered, and a normal status icon is displayed through the human-computer interaction interface; If the fault probability is within the interval between a preset first fault threshold and a preset second fault threshold, the speed regulation warning mechanism is triggered, a fault warning status icon is displayed through the human-computer interaction interface, and the feeding speed is adjusted to a preset safe speed or switched to a spare step distance; If the failure probability is greater than a preset second failure threshold, the emergency stop mechanism is triggered, a maintenance work order is displayed through the human-computer interaction interface, and the driving power supply of the feeding motor is disconnected.

7. A carrier material supply and demand forecasting system based on machine learning, characterized in that: The system includes a memory and a processor. The memory includes a carrier material supply and demand forecasting method program based on machine learning. When the carrier material supply and demand forecasting method program based on machine learning is executed by the processor, the following steps are implemented: According to the preset acquisition cycle, the feeding speed is read based on the preset instruction address, the motor current is collected based on the detection circuit, and the temperature information and vibration information are measured based on the preset environmental sensor; Generate a feeding speed sequence according to the feeding speed and sampling timestamp, and calculate and obtain preset rhythm standard deviation information; Based on a preset timing network model, the material remaining rate is obtained according to the feeding speed sequence and the rhythm standard deviation information; generating a material replenishment request instruction or a material replenishment warning instruction according to the relationship between the material remaining rate and the remaining rate threshold; generating a multi-source data sequence according to the motor current, temperature information, vibration information and timestamp; Obtaining a fault probability based on the multi-source data sequence based on a preset classifier model; Based on the preset hierarchical warning mechanism, the status display mechanism, speed regulation warning mechanism or emergency stop mechanism is triggered according to the failure probability.

8. The carrier material supply and demand forecasting system based on machine learning according to claim 7, characterized in that: The method includes: reading the feeding speed based on a preset instruction address according to a preset acquisition cycle, collecting the motor current based on a detection circuit, and measuring the temperature information and vibration information based on a preset environmental sensor. Get the instruction address information of feeding speed; Based on a preset communication protocol and according to the instruction address information, the instruction communication data is parsed to obtain the feeding speed; Based on the preset current sensor circuit, analog-to-digital information is converted by an analog-to-digital converter; Converting the motor current according to the analog-to-digital information; According to the preset temperature sensor, the real-time temperature value of the feeder is measured to obtain the temperature information; The vibration amplitude of the feeder motor is measured according to a preset vibration sensor to obtain the vibration information.

9. The carrier material supply and demand forecasting system based on machine learning according to claim 7, characterized in that: The calculation to obtain the preset rhythm standard deviation information specifically includes: According to the predicted duration of the timing network model, the feeding speed sequence is intercepted; Segmenting the intercepted feeding speed sequence according to a preset first time window; Calculate the speed average and speed variance of the feeding speed sequence in each window; Determine the speed tolerance based on the tape type; If the speed average value exceeds the speed allowable range, setting first identification information; If the speed variance exceeds a preset variance threshold, setting second identification information; The tempo standard deviation information is composed by combining the speed average, the speed variance, the first identification information, and the second identification information.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The computer-readable storage medium includes a carrier material supply and demand prediction method program based on machine learning. When the carrier material supply and demand prediction method program based on machine learning is executed by a processor, the steps of the carrier material supply and demand prediction method based on machine learning as described in any one of claims 1 to 6 are implemented.

Images