Equipment state monitoring method and system based on smart factory

Abstract

The invention discloses an equipment state monitoring method and system based on a smart factory. The equipment state monitoring method based on the smart factory comprises the following steps: acquiring equipment data; transmitting the acquired equipment data; processing the transmitted equipment data; constructing a multi-modal matrix according to the processed equipment data; inputting the multi-modal matrix into a learning network to carry out equipment state monitoring model training; and carrying out equipment state monitoring evaluation according to the equipment state monitoring model. A multi-modal sensor data acquisition strategy is adopted, equipment operation information is captured from multiple dimensions such as temperature, pressure, vibration and current, and limitation of single sensor data is avoided. And meanwhile, the collected data is quickly sent to the edge computing node through a wireless transmission module according to a specific protocol, and a data compression technology and a redundancy transmission mechanism are adopted, so that the transmission delay is reduced while the data integrity is ensured, and the monitoring real-time performance is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, in particular to a device state monitoring method and system based on a smart factory. BACKGROUND

[0002] In the production environment of a smart factory, stable operation of devices is crucial for production efficiency and product quality. Traditional device state monitoring methods have many shortcomings, for example, using a single sensor or a simple threshold detection method, which is difficult to fully capture the complex changes in the running state of the device, resulting in insufficient accuracy of fault prediction, and the maintenance decision is not scientific enough, which further causes frequent unplanned downtime and production loss. Especially in high-load production environments, the dynamic changes of device state are coupled with many factors, further increasing the difficulty of monitoring. Traditional fault prediction techniques mainly rely on deep neural networks and other algorithms, which have high prediction accuracy, but the network structure is complex and requires backpropagation iterative calculation, resulting in long model training time and large computing resource demand, making it difficult to deploy in real time on edge computing devices. In addition, in the data transmission process, traditional monitoring methods often cause data transmission delay or loss due to large data volume and non-uniform transmission protocols, affecting the real-time and accuracy of monitoring.

[0003] Therefore, how to provide a device monitoring method that can improve the reliability and production efficiency of the device has become a problem that needs to be solved in the field. SUMMARY

[0004] To solve the above problems, the present application provides a device state monitoring method based on a smart factory, comprising the following steps: obtaining device data; transmitting the obtained device data; processing the transmitted device data; constructing a multi-modal matrix according to the processed device data; inputting the multi-modal matrix into a learning network for device state monitoring model training; and performing device state monitoring evaluation according to the device state monitoring model.

[0005] The device state monitoring method based on a smart factory as described above, wherein obtaining device data comprises deploying sensors in the device and obtaining device data through the sensors.

[0006] The device state monitoring method based on a smart factory as described above, wherein the device data includes device temperature, pressure, vibration, and current data.

[0007] The device state monitoring method based on a smart factory as described above, wherein processing the transmitted device data comprises determining whether the device data is missing; if there is a missing value, the missing value is filled; and the missing data is represented as:

[0008]

[0009] x m-p x represents a valid data point before missing data, x m+q x represents a valid data point after missing data, m represents the position index of the missing data point in the data sequence, and p represents the interval number between the valid data point before the missing data point and the missing data point.

[0010] The device state monitoring method based on the smart factory as described above, wherein the multi-modal matrix comprises a plurality of modalities and features corresponding to each modality.

[0011] A device state monitoring system based on a smart factory comprises a device data acquisition unit, a transmission unit, a processing unit, a modality matrix construction unit, a model training unit, and a monitoring and evaluation unit. The device data acquisition unit is configured to acquire device data. The transmission unit is configured to transmit the acquired device data. The processing unit is configured to process the transmitted device data. The modality matrix construction unit is configured to construct a multi-modal matrix based on the processed device data. The model training unit is configured to input the multi-modal matrix into a learning network to train a device state monitoring model. The monitoring and evaluation unit is configured to perform device state monitoring and evaluation based on the device state monitoring model.

[0012] The device state monitoring system based on the smart factory as described above, wherein the device data acquisition unit acquires device data by deploying sensors in the device and acquiring device data through the sensors.

[0013] The device state monitoring system based on the smart factory as described above, wherein the device data acquired by the device data acquisition unit includes device temperature, pressure, vibration, and current data.

[0014] The device state monitoring system based on the smart factory as described above, wherein the processing unit processes the transmitted device data by determining whether the device data has missing data, and if so, performing data filling. The missing data is represented as:

[0015]

[0016] x m-p x represents a valid data point before missing data, x m+q x represents a valid data point after missing data, m represents the position index of the missing data point in the data sequence, and p represents the interval number between the valid data point before the missing data point and the missing data point.

[0017] The device state monitoring system based on the smart factory as described above, wherein the modality matrix construction unit constructs a multi-modal matrix comprising a plurality of modalities and features corresponding to each modality.

[0018] The present application has the following advantages:

[0019] (1) The application adopts a multi-modal sensor data acquisition strategy to capture device operation information from multiple dimensions such as temperature, pressure, vibration, current, etc., avoiding the limitations of single sensor data. Through iterative filtering interpolation algorithm for data smoothing processing, effectively eliminating noise interference, combined with improved extreme value detection algorithm to extract key feature points, providing high-quality input data for the model. At the same time, the jointly trained generalized Siamese-like extensive learning network can deeply mine the internal relationship between multi-modal data, even in the case of labeled data scarcity, still can maintain high learning ability, greatly improve the accuracy of device state evaluation and the precision of fault prediction.

[0020] (2) The data collected by the sensor is transmitted to the edge computing node quickly through the wireless transmission module with a specific protocol, and the data compression technology and redundant transmission mechanism are adopted to reduce the transmission delay while ensuring the data integrity, ensuring the real-time of monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0022] Figure 1 is a flow diagram of a device state monitoring method based on a smart factory provided by an embodiment of the present application;

[0023] Figure 2 is a schematic diagram of the internal structure of a device state monitoring system based on a smart factory provided by an embodiment of the present application. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all 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.

[0025] Embodiment one

[0026] As shown in Figure 1 , the present embodiment provides a device state monitoring method based on a smart factory, which specifically includes the following steps:

[0027] Step S1: Obtain device data.

[0028] In the smart factory, various types of sensors are deployed to acquire device data through sensors. The various types of sensors include temperature sensors, pressure sensors, vibration sensors, current sensors, humidity sensors, displacement sensors, etc. The various types of sensors collect device data such as temperature, pressure, vibration, current, humidity, and displacement. Temperature, pressure, vibration, current, humidity, and displacement are multiple data types in the device data.

[0029] According to the characteristics and monitoring requirements of different devices, sensors are distributed at key parts of each device. The specific deployment positions are as follows:

[0030] Core components of production equipment: For rotating machinery (such as motors, pumps, fans, etc.), vibration sensors are installed at vibration-sensitive parts such as bearing seats and machine housings to monitor the vibration of the equipment during operation and promptly detect faults such as bearing wear and rotor imbalance. Temperature sensors are placed close to motor windings and bearing outer rings to monitor the operating temperature of the equipment and prevent damage caused by overheating.

[0031] Transmission and connection parts: Displacement sensors are installed at the shaft ends of the drive wheels and driven wheels of the conveyor belt to monitor the deviation of the conveyor belt. Torque sensors are installed at the input shaft and output shaft of the gearbox to monitor the transmission torque in real time and determine the gear engagement.

[0032] Energy supply system: Current and voltage sensors are installed at locations such as distribution boxes and cable joints to monitor power parameters and promptly detect power faults such as short circuits and overloads. Pressure and flow sensors are installed before and after valves in steam pipes and compressed air pipes to monitor energy delivery status.

[0033] Environmental monitoring area: Temperature and humidity sensors are installed at the top and corners of the workshop to monitor the overall environmental temperature and humidity of the workshop and provide a basis for equipment operation environment control. Dust sensors are installed in areas with high dust levels to ensure production environment safety and equipment service life.

[0034] Material handling link: Liquid level sensors are installed at the bottom or sidewall of containers such as silos and storage tanks to monitor the storage amount of materials. Concentration sensors are installed in material conveying pipelines to monitor changes in material concentration and ensure stable production processes.

[0035] In summary, the acquired device data includes device temperature, pressure, vibration, and current data.

[0036] Step S2: Transmit the acquired device data.

[0037] The data collected by the sensor is transmitted to the edge computing node through the wireless transmission module according to a specific protocol, such as the MQTT protocol, the CoAP protocol, etc. For some key data, a redundant transmission mechanism is adopted, i.e., the data is transmitted through two different transmission paths at the same time to prevent data loss. Let the device data collected by the sensor be D = {d1, d2,... d i ,...d n}, where d i represents the data collected by the i-th sensor.

[0038] Preferably, during data transmission, a data compression technique such as a compression algorithm based on wavelet transform can be used to reduce the amount of data transmission and improve transmission efficiency.

[0039] Step S3: processing the transmitted device data.

[0040] In this embodiment, the edge computing node performs preliminary processing on the received data. First, data integrity check is performed by comparing the checksum of the received data with the checksum in the data packet. If they are inconsistent, the sensor is required to resend the data.

[0041] The device data collected by the sensor is smoothed to effectively eliminate random noise interference and retain the trend of device state changes. Specifically, an adaptive weighted moving average method is used to determine the weighted moving average value. The weighted moving average value is calculated by assigning different weights to the data within a certain window and calculating the average value, which can effectively smooth these random fluctuations and make the data sequence more stable, better reflecting the true trend of the device running state.

[0042] After smoothing, it is also necessary to confirm whether the obtained device data has data missing.

[0043] If there is data missing, data filling is performed. Let the missing data be x m , the previous and next valid data points be x m-p and x m+q , and the missing data be represented as:

[0044]

[0045] m represents the position index of the missing data point in the data sequence; p represents the interval number between the previous valid data point and the missing data point, i.e., the previous valid data point is located p positions before the missing data point. For example, if the missing data point is the 10th data and the previous valid data point is the 7th data, then m = 10 and p = 3.

[0046] Step S4: constructing a multi-modal matrix according to the processed device data.

[0047] It can be understood that the multi-modal matrix includes a plurality of features corresponding to a plurality of modalities. Each of the modalities corresponds to a certain type of device data collected, for example, the temperature modality corresponds to temperature data collected by a temperature sensor, and the current modality corresponds to current data collected by a current sensor.

[0048] The feature corresponding to each modality is representative information extracted after processing and analyzing the original device data under the modality. For example, for the temperature modality (i is the index corresponding to the temperature modality), the jth feature can be the average temperature, which is obtained by averaging the original temperature data collected by the temperature sensor within a period of time. Still as an example, the current peak feature under the current modality is the maximum value extracted from the original current data collected by the current sensor.

[0049] It can be understood that the above features can be determined according to the type of the device and the monitoring target. For example, for a motor device, the features of the current modality can include the current effective value, the current peak value, the current harmonic content, etc.; and the features of the temperature modality can include the average temperature, the maximum temperature, the temperature change rate, etc.

[0050] In summary, assuming that the processed device data is divided into m modalities, and each modality has n features, the multi-modal feature matrix M is represented as:

[0051]

[0052] where f ij represents the jth feature of the ith modality, i = 1, 2,..., m, j = 1, 2,..., n.

[0053] Further, the features are normalized to unify different orders of magnitude to the same value range, such as [0, 1], and the jth feature f ij ' of the ith modality after normalization is represented as:

[0054] f ij ' = f ij -min(f i ) / max(f i )-min(f i )

[0055] where min(f i ) represents the minimum value of the ith modality feature, and max(f i ) represents the maximum value of the ith modality feature.

[0056] In order to ensure the smooth construction of the multi-modal matrix, it is further included that the device data obtained is stored in the edge node before the multi-modal matrix is constructed.

[0057] The device data is divided into multiple device data blocks, part of the device data blocks are stored to the first gradient edge node, part of the device data blocks are stored to the second gradient edge node, and part of the device data blocks are stored to the third gradient edge node. When the first to the third gradient cannot store all the multiple data blocks, the remaining data blocks are stored to the cloud node.

[0058] The storage energy consumption E of the edge node is represented as:

[0059]

[0060] ρ n→k represents the data amount of the device data block n stored to the kth gradient edge node, C n,k represents the CPU revolutions required when storing the device data block n in the kth gradient edge node, m n,k represents the energy consumed by the CPU per revolution when storing the device data block n in the kth gradient edge node, t n,k represents the time required when storing the device data block n in the kth gradient edge node, P represents the transmission loss of the device data block n transmitted to the kth gradient edge node, K represents the total number of gradients, and N represents the total amount of device data.

[0061] When the storage energy consumption E is greater than a specified threshold, the remaining device data blocks are stored to the cloud node, otherwise the remaining device data blocks are randomly stored in the kth gradient.

[0062] It can be understood that each gradient contains one or more edge nodes. When the device data block n is stored in multiple edge nodes in the kth gradient, the above parameters are calculated based on the maximum execution value, for example, when the device data block n is stored in three edge nodes in the first gradient, the CPU revolutions of the three edge nodes when executing are 2k, 3k, and 4k, then C n,k takes 4k.

[0063] Step S5: inputting the multi-modal matrix into the learning network to train the device state monitoring model.

[0064] A joint training generalized Siamese-like extensive learning network is constructed, and the multi-modal matrix is inputted into the Siamese-like extensive learning network to train the model. Step S5 includes the following sub-steps:

[0065] Step S51: determining the learning network.

[0066] A joint training generalized Siamese-like wide learning network is constructed, which is composed of an input layer, a feature mapping layer, an enhancement layer and an output layer. The input layer receives a multi-modal feature matrix, the feature mapping layer maps the input features to a high-dimensional feature space through random weights, the enhancement layer performs nonlinear transformation on the high-dimensional features, and the output layer outputs the device state evaluation result and the fault prediction probability.

[0067] Step S52: input the multi-modal matrix into the learning network for model training.

[0068] The multi-modal matrix is input into the deep learning network for model training to obtain the device state monitoring model.

[0069] During the training process, the data set is divided into a training set, a validation set and a test set in a batch training manner, and the proportion is 7:2:1. Let the loss function of the network be L, which is represented as:

[0070]

[0071] Where N is the number of samples, y i is the true label, is the predicted label.

[0072] The network parameters are adjusted by optimizing the loss function, such as using the stochastic gradient descent algorithm, and the parameter update formula is:

[0073] θ t+1 = θ t - α▽L(θ t )

[0074] θ t+1 is the updated parameter, θ t is the current parameter, α is the learning rate, the initial learning rate is set to 0.01, which gradually decays with the increase of the number of training times, and the decay coefficient is 0.95,▽L(θ t ) is the gradient of the loss function L at the current parameter.

[0075] When the loss function of the validation set no longer decreases continuously for a specified number of iterations, the training is stopped, and the current model is saved as the device state monitoring model.

[0076] Step S6: device state monitoring and evaluation according to the device state monitoring model.

[0077] The device state monitoring model performs deep analysis and learning on the input multi-modal feature matrix, can accurately mine the potential laws and state characteristics hidden in the device operation data, and thus realizes the evaluation of the current running state of the device and the early prediction of the possible future faults.

[0078] The device state monitoring model outputs the average failure-free operation time of the device and the comprehensive efficiency of the device through the input multi-modal matrix. The greater the average failure-free operation time of the device, the higher the reliability of the device. The higher the comprehensive efficiency of the device, the higher the effective utilization rate of the device within the planned operation time.

[0079] It can be understood that the comprehensive efficiency of the device needs to be determined according to device data such as time utilization rate, production quantity and qualified product quantity, and is specifically represented as:

[0080] K = (T 实际运行 ÷ T 计划运行 ) × (C 实际产量 ÷ (T 实际运行 × S 理论生产 ) × R 合格 ) × 100%

[0081] Wherein K represents the comprehensive efficiency of the device, T 实际运行 represents the actual operation time, T 计划运行 represents the planned operation time, C 实际产量 represents the actual output, S 理论生产 represents the theoretical production rate, and R 合格 represents the qualified product rate.

[0082] Example two

[0083] As shown in Figure 2 , a device state monitoring system based on a smart factory is provided for the embodiments of the present application, and specifically includes: a device data acquisition unit 210, a transmission unit 220, a processing unit 230, a modal matrix construction unit 240, a model training unit 250 and a monitoring and evaluation unit 260.

[0084] The device data acquisition unit 210 is used to acquire device data.

[0085] A plurality of types of sensors are deployed in the smart factory, and device data is acquired through the sensors. The plurality of types of sensors include temperature sensors, pressure sensors, vibration sensors, current sensors, humidity sensors, displacement sensors, etc. The plurality of types of sensors collect temperature, pressure, vibration, current, humidity and displacement device data. Temperature, pressure, vibration, current, humidity and displacement are a plurality of data types in the device data.

[0086] According to the characteristics and monitoring requirements of different devices, they are distributed at key positions of each device. The specific deployment positions are as follows:

[0087] Core components of production equipment: For rotating machinery (such as motors, pumps, fans, etc.), vibration sensors are installed in vibration-sensitive parts such as bearing seats and machine housings to monitor the vibration during equipment operation and timely detect bearing wear, rotor imbalance and other faults; temperature sensors are attached to motor windings, bearing outer rings and other parts to monitor equipment operating temperature and prevent equipment damage due to overheating.

[0088] Transmission and connection parts: Displacement sensors are installed on the drive wheel and driven wheel shaft end of the conveyor belt to monitor the deviation of the conveyor belt. Torque sensors are installed on the input shaft and output shaft of the gearbox to monitor the transmission torque in real time and determine the gear engagement.

[0089] Energy supply system: Current and voltage sensors are installed at the distribution box and cable joint to monitor power parameters and detect power faults such as short circuit and overload in a timely manner. Pressure and flow sensors are installed before and after the valve of the steam pipeline and compressed air pipeline to monitor the energy delivery status.

[0090] Environmental monitoring area: Temperature and humidity sensors are installed at the top and corners of the workshop to monitor the overall environmental temperature and humidity of the workshop and provide a basis for equipment operation environment control. Dust sensors are installed in areas with high dust to ensure the safety of the production environment and the service life of the equipment.

[0091] Material handling link: Liquid level sensors are installed at the bottom or sidewall of silos, tanks and other containers to monitor the storage capacity of materials; concentration sensors are installed in material conveying pipelines to monitor the concentration changes of materials and ensure the stability of the production process.

[0092] In summary, the obtained equipment data includes equipment temperature, pressure, vibration and current data.

[0093] The transmission unit 220 is configured to transmit the obtained equipment data.

[0094] The data collected by the sensors is transmitted to the edge computing node through the wireless transmission module according to a specific protocol, such as the MQTT protocol or the CoAP protocol. For some critical data, a redundant transmission mechanism is used, i.e., the data is transmitted through two different transmission paths at the same time to prevent data loss. Let the equipment data collected by the sensors be D = {d1, d2,... d i ,...d n}, where di represents the data collected by the ith sensor. i

[0095] Preferably, during data transmission, data compression techniques such as wavelet transform-based compression algorithms can be used to reduce data transmission volume and improve transmission efficiency.

[0096] ​The processing unit 230 is configured to process the device data transmitted.

[0097] The edge computing node performs preliminary processing on the received data, first performs data integrity check, and compares the checksum of the received data with the checksum in the data packet. If they are inconsistent, the sensor is required to resend the data.

[0098] The device data collected by the sensor is smoothed to effectively eliminate random noise interference and retain the trend of device state change. Specifically, the adaptive weighted moving average method is used to determine the weighted moving average value. The weighted moving average value is obtained by assigning different weights to the data in a certain window and calculating the average value, which can effectively smooth the random fluctuations and make the data sequence more stable and better reflect the real trend of the device running state.

[0099] After smoothing, it is further determined whether the obtained device data is missing.

[0100] If there is data missing, the data is filled. Let the missing data be x m , the previous and next valid data points are x m-p and x m+q , and the missing data is represented as:

[0101]

[0102] m represents the position index of the missing data point in the data sequence; p represents the interval number between the previous valid data point and the missing data point, i.e. the previous valid data point is located at the pth position before the missing data point. For example, if the missing data point is the 10th data, and the previous valid data point is the 7th data, then m=10 and p=3.

[0103] The modal matrix construction unit 240 is configured to construct a multi-modal matrix according to the processed device data.

[0104] It can be understood that the multi-modal matrix includes a plurality of features corresponding to a plurality of modalities. Each modality in the multi-modal matrix corresponds to a certain type of device data collected, for example, the temperature modality corresponds to the temperature data collected by the temperature sensor, and the current modality corresponds to the current data collected by the current sensor.

[0105] Each feature corresponding to each modality is representative information extracted from the original equipment data under the modality after processing and analysis. As an example, for the temperature modality (i is the index corresponding to the temperature modality), the jth feature can be the average temperature, which is obtained by averaging the original temperature data collected by the temperature sensor within a period of time. Still as an example, the current peak value feature under the current modality is the maximum value extracted from the original current data collected by the current sensor.

[0106] It can be understood that the above features can be determined according to the type of the equipment and the monitoring target. For example, for a motor equipment, the features of the current modality can include the current effective value, the current peak value, the current harmonic content, etc.; the features of the temperature modality can include the average temperature, the maximum temperature, the temperature change rate, etc.

[0107] In summary, assuming that the processed equipment data is divided into m modalities, and each modality has n features, the multi-modality feature matrix M is represented as:

[0108]

[0109] where f ij represents the jth feature of the ith modality.

[0110] Further, the features are normalized to unify the features of different magnitudes to the same value range, such as [0, 1], and the jth feature f ij ' of the ith modality after normalization is represented as:

[0111] f ij ' = f ij -min(f i ) / max(f i )-min(f i )

[0112] where min(f i ) represents the minimum value in the ith modality feature, and max(f i ) represents the maximum value in the ith modality feature.

[0113] In order to ensure the smooth construction of the multi-modality matrix, it is further included that before constructing the multi-modality matrix, the obtained equipment data is stored in the edge node.

[0114] The equipment data is divided into a plurality of equipment data blocks, part of the equipment data blocks are stored in the first gradient edge node, part of the equipment data blocks are stored in the second gradient edge node, and part of the equipment data blocks are stored in the third gradient edge node. When the first to third gradients cannot store all the plurality of data blocks, the remaining data blocks are stored in the cloud node.

[0115] wherein the storage energy consumption E of the edge node is expressed as:

[0116]

[0117] ρ n→k represents the data volume of the device data block n stored to the kth echelon edge node, C n,k represents the CPU revolutions required when storing the device data block n at the kth echelon edge node, m n,k represents the energy consumed per revolution of the CPU when storing the device data block n at the kth echelon edge node, t n,k represents the time required when storing the device data block n at the kth echelon edge node, P represents the transmission loss of the device data block n transmitted to the kth echelon edge node, K represents the total number of echelons, and N represents the total device data volume.

[0118] When the storage energy consumption E is greater than a specified threshold, the remaining device data block is stored to the cloud node, otherwise the remaining device data block is randomly stored in the kth echelon.

[0119] It can be understood that each echelon contains one or more edge nodes. When the device data block n is stored in multiple edge nodes in the kth echelon, the above parameters are calculated based on the maximum execution value, for example, when the device data block n is stored in 3 edge nodes in the first echelon, the CPU revolutions of the 3 edge nodes when executing are 2k, 3k, and 4k, then C n,k takes 4k.

[0120] The model training unit 250 is configured to input the multi-modal matrix into a learning network to train a device state monitoring model.

[0121] The model training unit 250 performs the following steps:

[0122] Step T1: Determine the learning network.

[0123] A joint training generalized Siamese-like extensive learning network is constructed, which is composed of an input layer, a feature mapping layer, an enhancement layer, and an output layer. The input layer receives a multi-modal feature matrix, the feature mapping layer maps the input features to a high-dimensional feature space through random weights, the enhancement layer performs nonlinear transformation on the high-dimensional features, and the output layer outputs a device state evaluation result and a fault prediction probability.

[0124] Step T2: Input the multi-modal matrix into the learning network to train the model.

[0125] The multi-modal matrix is input into the deep learning network to train the model, and a device state monitoring model is obtained.

[0126] In the training process, the data set is divided into training set, validation set and test set in batch training mode, and the proportion is 7:2:1. The loss function of the network is L, which is represented as:

[0127]

[0128] Where N is the number of samples, y i is the true label, is the predicted label.

[0129] The network parameters are adjusted by optimizing the loss function, such as using the stochastic gradient descent algorithm, and the parameter update formula is:

[0130] θ t+1 = θ t - α▽L(θ t )

[0131] θ t+1 is the updated parameter, θ t is the current parameter, α is the learning rate, the initial learning rate is set to 0.01, and it gradually decays with the increase of the number of training times, and the decay coefficient is 0.95,▽L(θ t ) is the gradient of the loss function L under the current parameter.

[0132] When the loss function of the validation set does not decrease continuously for a specified number of iterations, the training is stopped, and the current model is saved as the device state monitoring model.

[0133] The monitoring and evaluation unit 260 is used to monitor and evaluate the device state according to the device state monitoring model.

[0134] The device state monitoring model performs deep analysis and learning on the input multi-modal feature matrix, can accurately mine the potential laws and state characteristics contained in the device running data, and thus realize the evaluation of the current running state of the device and the early prediction of the possible future failure.

[0135] The device state monitoring model outputs the average failure-free running time of the device and the comprehensive efficiency of the device through the input multi-modal matrix. The larger the average failure-free running time value of the device is, the higher the reliability of the device is. The higher the comprehensive efficiency value of the device is, the higher the effective utilization rate of the device in the planned running time is.

[0136] The application also provides a computer storage medium, which stores computer instructions, and the computer instructions are used to execute the device state monitoring method based on the smart factory when called.

[0137] The disclosed embodiment provides a computer readable storage medium, which stores computer program instructions, and when the computer program instructions are run on a computer, the computer executes the above-mentioned device state monitoring method based on a smart factory.

[0138] The embodiment of the present application provides a processor for processing the above-mentioned device state monitoring method based on a smart factory.

[0139] In the embodiment of the present application, the processor can be an integrated circuit chip with signal processing capability. The processor can be a general processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0140] The disclosed methods, steps and logic block diagrams in the embodiments of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can be any conventional processor. The steps of the method disclosed in combination with the embodiments of the present application can be directly embodied as hardware code processor execution or executed by hardware and software module combination in code processor. The software module can be located in the random storage, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, register and other mature storage medium in the art. The processor reads the information in the storage medium and combines the hardware to complete the steps of the above method.

[0141] The storage medium can be a memory, for example, can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories.

[0142] The non-volatile memory can be a Read-Only Memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an Electrically EPROM (EEPROM), or a flash memory. The volatile memory can be a Random Access Memory (RAM) used as an external cache. By way of example, and not limitation, many forms of RAM can be used, such as a Static RAM (SRAM), a Dynamic RAM (DRAM), a Synchronous DRAM (SDRAM), a Double Data Rate SDRAM (DDR SDRAM), an Enhanced SDRAM (ESDRAM), a Synchlink DRAM (SLDRAM), and a Direct Rambus RAM (DRRAM).

[0143] The present application has the following beneficial effects:

[0144] (1) The present application adopts a multi-modal sensor data acquisition strategy to capture equipment operation information from multiple dimensions such as temperature, pressure, vibration, and current, avoiding the limitations of single sensor data. Through iterative filtering interpolation algorithm for data smoothing processing, effectively eliminating noise interference, combined with improved extreme value detection algorithm to extract key feature points, providing high-quality input data for the model. At the same time, the jointly trained generalized Siamese-like extensive learning network can deeply mine the internal correlation between multi-modal data, even in the case of labeled data scarcity, still can maintain high learning ability, greatly improves the accuracy of equipment state evaluation and the precision of fault prediction.

[0145] (2) The data collected by the sensor of the present application is quickly sent to the edge computing node through the wireless transmission module with a specific protocol, adopts data compression technology and redundant transmission mechanism, reduces the transmission delay while ensuring the integrity of the data, ensures the real-time of monitoring.

[0146] Although the examples referred to in the present application are described, which are only for the purpose of explanation and not limitation of the present application, changes, additions and / or deletions to the embodiments can be made without departing from the scope of the present application.

[0147] The above description is only specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for monitoring equipment status in a smart factory, characterized in that, Includes the following steps: Acquire device data; Transmit the acquired device data; Process the transmitted device data; Construct a multimodal matrix based on the processed device data; The multimodal matrix is ​​input into the learning network to train the device status monitoring model; Equipment status monitoring and assessment are conducted based on the equipment status monitoring model.

2. The equipment status monitoring method based on a smart factory as described in claim 1, characterized in that, Acquiring device data includes deploying sensors in the device and acquiring device data through the sensors.

3. The equipment status monitoring method based on a smart factory as described in claim 2, characterized in that, Equipment data includes equipment temperature, pressure, vibration, and current data.

4. The equipment status monitoring method based on a smart factory as described in claim 3, characterized in that, Processing the transmitted device data includes determining whether the device data is missing; If any data is missing, it will be filled in. Missing data is represented as: x m-p x represents the valid data point before the missing data. m+q This represents the valid data point after the missing data point, where m represents the index of the missing data point in the data sequence, and p represents the number of intervals between the previous valid data point and the missing data point.

5. The equipment status monitoring method based on a smart factory as described in claim 4, characterized in that, A multimodal matrix includes multiple modes and the features corresponding to each mode.

6. A smart factory-based equipment status monitoring system, characterized in that, include: The equipment includes a data acquisition unit, a transmission unit, a processing unit, a modality matrix construction unit, a model training unit, and a monitoring and evaluation unit. Equipment data acquisition unit, used to acquire equipment data; The transmission unit is used to transmit the acquired device data; The processing unit is used to process the transmitted device data; The modality matrix construction unit is used to construct a multimodal matrix based on the processed device data. The model training unit is used to input the multimodal matrix into the learning network to train the device status monitoring model; The monitoring and evaluation unit is used to monitor and evaluate the equipment status based on the equipment status monitoring model.

7. The equipment status monitoring system based on a smart factory as described in claim 6, characterized in that, The device data acquisition unit acquires device data by deploying sensors in the device and acquiring device data through the sensors.

8. The equipment status monitoring system based on a smart factory as described in claim 7, characterized in that, The equipment data acquired by the equipment data acquisition unit includes equipment temperature, pressure, vibration, and current data.

9. The equipment status monitoring system based on a smart factory as described in claim 8, characterized in that, The processing unit processes the transmitted device data, including determining whether the device data is missing; If any data is missing, it will be filled in. Missing data is represented as: x m-p x represents the valid data point before the missing data. m+q This represents the valid data point after the missing data point, where m represents the index of the missing data point in the data sequence, and p represents the number of intervals between the previous valid data point and the missing data point.

10. The equipment status monitoring system based on a smart factory as described in claim 9, characterized in that, The modality matrix constructed by the modality matrix building unit includes multiple modes and the features corresponding to each mode.

Images