A multi-dimensional pressure vessel health monitoring system and method based on artificial intelligence on-site calculation

Abstract

The present invention relates to a multi-dimensional pressure vessel health monitoring system and method based on artificial intelligence on-site calculation, wherein N types of sensors are installed on the pressure vessel, and the parallel real-time data obtained by detection are uniformly adapted to obtain parallel data streams, and a sampling data matrix is ​​obtained through storage and sliding window processing. As the window slides, the change process of the eigenvalue of the window data matrix is ​​correlated with the characteristic model of different health states, and the likelihood ratio between the real-time data model and various health state models can be obtained. After the likelihood ratio value is accumulated with a length of L samples, a state judgment is performed, and a monitoring state information data stream is output, and then the data volume is encoded to form transmission data, which is sent to the monitoring system platform through a transmission channel. The present invention realizes multi-dimensional joint monitoring of multiple different types of sensor data, and has the characteristics of high sensitivity, low false alarm rate, low network traffic dependence and low power consumption.

Description

Technical Field

[0001] The present invention relates to the technical field of pressure vessel health detection technology, and specifically, to a multi-dimensional sensor data joint monitoring method and system implemented through artificial intelligence on-site calculation. Background Art

[0002] Pressure vessels for storing and transporting hazardous chemicals, especially those used for transporting hazardous chemicals, are widely distributed in public places such as urban traffic hubs, factories, hotels, guesthouses, and residential buildings. They are also mobile, with many risk management factors and difficult management. Once an accident occurs, it often endangers public safety, causing huge losses and social impact. Monitoring the health status of these hazardous chemical pressure vessels is the primary link in safety management and accident warning. At present, the monitoring of the health status of pressure vessels mainly uses various discrete sensors such as pressure, liquid level, temperature, and gas detection for real-time data collection, and then uploads the sensor values ​​to the monitoring system platform through various Internet of Things networks for storage and real-time health status judgment. The monitoring system platform issues abnormal alarms based on changes in the values ​​collected by various sensors. For example, the patent application document with application number CN202110572980.2 discloses an online monitoring system for the operation status of a boiler pressure vessel. The online monitoring data acquisition module collects the traditional wall thickness, and further adopts acoustic emission, high temperature strain testing, deformation monitoring and other means to collect the thickness change, crack initiation, local stress state, vibration and tilt degree of the equipment in real time during operation. The online monitoring data display module presents it, and sends the visualized operation parameters to the health monitoring and fault warning system platform in real time. The online monitoring data alarm module is used to calculate and process the collected data, and then compares it with the set threshold to evaluate the health status of major high temperature pressure equipment in real time, so that users can grasp the operation status of the current equipment in real time, predict the occurrence of faults in time, and provide strong protection for the user's safe production. The present invention has the advantages of boiler pressure vessel operation status monitoring, operation status visualization, and timely alarm to achieve the effect of convenient user use.

[0003] The patent application document with application number CN202110683934.X discloses a pressure vessel leakage detection system and its working method, including a central processing unit, an information acquisition unit, a data analysis unit, a data processing unit, a database, a monitoring and positioning unit, an alarm unit and a transmission unit. The data output end of the information acquisition unit is connected to the data receiving end of the data analysis unit and the database. The data analysis unit transmits the analyzed data results to the data processing unit, and the data processing unit transmits the processing results to the central processing unit. The alarm unit alarms according to the data information fed back by the central processing unit. It uses the information acquisition unit to monitor various data of the pressure vessel in real time, and uses the data analysis unit, data processing unit, database, monitoring and positioning unit, and alarm unit connected to the central processing unit for analysis and judgment, and formulates a treatment plan, which can timely discover abnormalities of the pressure vessel, effectively control the occurrence of accidents, and increase safety.

[0004] At present, the monitoring of the health status of pressure vessels usually monitors each type of sensor data separately. The conditions for triggering the status alarm also rely only on the simple sensor value change rate or the preset threshold comparison judgment. In addition, the existing monitoring methods require all sensor data to be uploaded to the system platform server. Since the environment of hazardous chemical vehicles is very complex during transportation and storage, the single pressure, liquid level, temperature, and gas detection values ​​change greatly. The alarm trigger strategy for the time-varying value or threshold comparison of a single type of sensor data has a high probability of false alarm. In order to reduce the probability of false alarms caused by individual performance differences of sensors in the above system, some systems currently use multiple parallel sensors, and use voting algorithms on the system platform to make alarm judgments for these parallel sensor values. The above detection system using multiple parallel sensor value voting algorithms not only increases the equipment cost, but also causes the IoT data traffic to increase several times. However, because the various types of sensor data are still judged independently, the false alarm problem of a single sensor system in a complex transportation environment has not been solved, and the probability of false alarms is reduced very limitedly. Summary of the invention

[0005] In order to overcome the shortcomings of the existing pressure vessel health monitoring system for hazardous chemical vehicles, the main purpose of the present invention is to provide a multi-dimensional pressure vessel health monitoring system and method based on artificial intelligence on-site calculation, and a multi-dimensional joint monitoring method of various types of sensors (pressure, liquid level, temperature, gas...) based on artificial intelligence on-site calculation, which has the advantages of high sensitivity, low false alarm rate and low network traffic dependence.

[0006] To solve the above technical problems, the present invention is achieved as follows:

[0007] A multi-dimensional pressure vessel health monitoring system based on artificial intelligence on-site computing includes an artificial intelligence on-site computing platform, which is composed of an on-site Internet of Things interface, a sensor data access module, a computing and storage module, and a mobile network access module.

[0008] The on-site IoT interface provides corresponding interfaces according to the communication modes of the N types of pressure vessel sensors to be connected, and automatically obtains parallel real-time data collected by N sensors;

[0009] The sensor can be a pressure sensor, a temperature sensor, a gas sensor, or a wind speed sensor, and can be installed at different locations of the pressure vessel as required.

[0010] Specifically, for tank trucks, pressure sensors are installed on the diagonals of the vehicle body, temperature sensors are installed on the outer cylinder of the tank truck, and wind speed sensors are installed in the parking spaces.

[0011] The access can be 485 serial port, and also includes 232 serial port, Ethernet, I²C field bus, MODBUS bus, and existing Internet of Things access methods such as LORA wireless and ZigBee wireless.

[0012] The sensor data access module samples and normalizes the acquired parallel real-time data according to the requirements of the multi-dimensional joint monitoring module, and then uniformly encodes it; after uniform encoding, a continuous N-dimensional data stream is formed. Through the sensor data access module of the artificial intelligence field computing platform, various types of sensor data are mapped into a unified multi-dimensional data space, so that the method proposed in the present invention has the basis for correlation judgment of various types of sensor values.

[0013] The real-time data normalization includes time correction and sampling rate unification.

[0014] The calculation and storage module stores the above N-dimensional data stream in real time, and uses the multi-dimensional joint monitoring module to perform model processing on the N-dimensional data stream to form a monitoring status information data stream, and stores the obtained monitoring status information data stream in real time and sends it to the mobile network access module. Through the calculation and storage module, a large amount of sensor collected data is calculated through correlation statistics, and the amount of status information data formed is less than one percent of the original sensor data, which reduces the demand for mobile communication network quality and bandwidth, and greatly improves the real-time monitoring and reliability of the method proposed by the present invention.

[0015] The mobile network access module establishes network data transmission channels according to the requirements of different types of wide area network access, encodes the monitoring status information data stream sent by the computing and storage module to form transmission data, and sends the transmission data to the monitoring system platform through the established transmission channel. This module encodes the monitoring status data, improving the reliability of data transmission on the wide area network and the security of information.

[0016] The multi-dimensional joint monitoring module includes a data window control module, a data model building module, a feature model training module, a real-time model likelihood ratio calculation module, and a real-time monitoring state judgment module, wherein:

[0017] The data window control module controls the received N-dimensional data stream to slide in the algorithm window according to the set step length according to the sampling rate and the algorithm window length L, and submits the N×L-dimensional data matrix in each sliding window to the multi-dimensional joint monitoring algorithm data model building module, such as Figure 3 As shown in the figure, the algorithm window L ≥ N, and the selection of the specific calculation window length depends on the system's balance between monitoring accuracy and computing platform computing power.

[0018] The data model building module forms a real-time window data model by calculating the data space feature vector according to the N×L dimensional data matrix submitted by the data window control module, and submits the real-time window data model to the real-time model likelihood ratio calculation module.

[0019] The real-time model likelihood ratio calculation module calculates the model space likelihood ratio of the received real-time window data model and the feature model, and simultaneously submits the likelihood ratio value to the real-time monitoring state judgment module and the feature model training module.

[0020] A real-time monitoring state judgment module, which uses the judgment criteria set by the multi-dimensional joint monitoring module to judge the likelihood ratio of the received current window data model, generates a current monitoring state data sequence of length N, and sends the monitoring state data sequence to the mobile network access module;

[0021] The feature model training module uses the feature model initialized and trained by the multi-dimensional joint monitoring module as the starting feature model, and corrects the local feature model according to the model spatial distribution when receiving the likelihood ratio value of each window data, so that the entire system of the present invention can continuously adapt to the judgment conditions of the environment and the state of repair, thereby improving the accuracy of the system alarm.

[0022] Beneficial effects: The present invention overcomes the problems of existing health monitoring systems for pressure vessels used in hazardous chemical vehicles, which rely on single sensor data, remote platform judgment, and mobile network data bandwidth. By using artificial intelligence calculations on the pressure vessel site, it realizes multi-dimensional joint monitoring of multiple different types of (pressure, liquid level, temperature, gas...) sensor data, and has the characteristics of high sensitivity, low false alarm rate, low network traffic dependence and low power consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a schematic diagram of the working principle of the multi-dimensional pressure vessel health monitoring system based on artificial intelligence on-site calculation of the present invention;

[0024] Figure 2 It is a composition diagram of the multi-dimensional pressure vessel health monitoring module;

[0025] Figure 3 is a schematic diagram of the principle of the data window control module performing sliding window processing on the continuous parallel data stream in step S3;

[0026] Figure 4 This is a schematic diagram of the sensor layout of an LNG tanker.

[0027] Among them, 1, 2-tank truck pressure sensors, 3-gas sensor components, 4-wind speed sensor, 5-tank truck outer cylinder temperature sensor, 6-on-site artificial intelligence monitoring system equipment. DETAILED DESCRIPTION

[0028] The present invention is further described in detail below in conjunction with specific implementation modes.

[0029] Example 1

[0030] This embodiment takes the implementation of LNG liquefied natural gas tanker health monitoring as an example to illustrate the multi-dimensional pressure vessel health monitoring method of artificial intelligence on-site calculation of the present invention.

[0031] like Figure 4 As shown, the left side is the front of the LNG transport truck, and the right side is the rear of the truck. The tank truck is equipped with a total of 5 sensors, namely, tank truck pressure sensors 1 and 2 installed on the right front and left rear of the vehicle body, gas sensor 3 installed inside the operation box at the rear of the vehicle, wind speed sensor 4 outside the operation box at the rear of the vehicle, and tank truck outer cylinder temperature sensor 5. The on-site artificial intelligence monitoring system 6 of the present invention is installed on the outer cylinder of the equipment tank truck. The 5-way sensor is connected to the on-site artificial intelligence monitoring system through one of the existing Internet of Things access methods such as 485 serial port, 232 serial port, Ethernet, I²C field bus, MODBUS bus, LORA wireless, ZigBee wireless, etc.

[0032] The data of the vehicle body and sensor equipment are as follows:

[0033] The maximum working pressure of the tank is 0.8Mpa, the design pressure is 1.0MPa, and it is equipped with 2 pressure sensors (tank truck pressure sensors 1 and 2) with a measuring range of 0.1~1.5Mpa and a sampling period of 1 second.

[0034] The outer cylinder of the tank is designed to work at a temperature of -20°C to 50°C. It is equipped with a temperature sensor (tank truck outer cylinder temperature sensor 5) with a range of -30°C to 70°C and a sampling period of 10 seconds.

[0035] Equipped with a gas sensor (gas sensor 3) with a measurement range of 1~99999 and a sampling period of 5 seconds.

[0036] Equipped with a wind speed transmitter (wind speed sensor 4) with a measuring range of 0~30m / s and a sampling period of 2 seconds.

[0037] Reference Figure 1 The processing flow of a multi-dimensional pressure vessel health monitoring method for artificial intelligence on-site calculation of the present invention includes the following steps:

[0038] Step S1, install the above 5 sensors at the corresponding positions of the LNG tanker as required, and access the on-site artificial intelligence monitoring system of the present invention through the 485 serial port. The actual access method is not limited to the 485 serial port, but also includes 232 serial port, Ethernet, I²C field bus, MODBUS bus, and existing Internet of Things access methods such as LORA wireless and ZigBee wireless. Various access data are first connected to the on-site Internet of Things interface of the artificial intelligence on-site computing platform, and the unified adaptation of various access forms and electrical interfaces is completed in this step.

[0039] Step S2, the sensor data access module of the artificial intelligence field computing platform aims to complete the unified adaptation of data collected by various sensors.

[0040] First, the different sampling periods of different sensors are unified. According to the shortest sampling period of each sensor, the data of other sensors are interpolated, which can be done by linear interpolation or digital filtering interpolation.

[0041] In this embodiment, taking the sampling period of the pressure sensor with the shortest sampling period as the standard, 1:9 interpolation is performed on the temperature sensor data, 1:4 interpolation is performed on the gas sensor data, and 1:1 interpolation is performed on the wind speed sensor data, so that all data have the same sampling period of 1 second, thereby achieving a unified sampling period for all sensor data.

[0042] Secondly, the sampling signal data of various sensors are normalized according to their range:

[0043] For the pressure sensor range of 0.1~1.5Mpa, perform linear normalization between 0~1;

[0044] The temperature sensor range is -30℃~70℃, and the linear normalization is performed between 0 and 1;

[0045] The gas sensor measurement range of 1~99999 is linearly normalized between 0~1;

[0046] Perform 0~1 linear normalization on the wind speed transmitter range of 0~30m / s.

[0047] In this way, all sensor data are evenly distributed between 0 and 1, and the time correction and sampling rate unification of all collected data are completed; thus, 5 parallel data streams are obtained.

[0048] Then, through the sensor data access module of the artificial intelligence field computing platform, various types of sensor data are mapped into a unified multi-dimensional data space, so that the method proposed in the present invention has the basis for associating and judging the values ​​of various types of sensors.

[0049] Step S3, the artificial intelligence on-site computing platform computing and storage module first stores the above 5 parallel data streams in real time. At the same time, the processing of the multi-dimensional joint monitoring module is adopted. Specifically, in the multi-dimensional joint monitoring algorithm data window control module, a data window L=5 sampling length is selected to perform sliding window processing on the continuous parallel data streams, such as Figure 3 As shown in the figure, a 5×5 sampled data matrix is ​​obtained in the algorithm window. If the computing power of the platform allows, the window sliding step size can be 1, that is, the data matrix is ​​processed once each sampling point is slid.

[0050] In the actual tank truck environment, there is a correlation between pressure, temperature, gas detection and wind speed data. For example, in the absence of leakage, an increase in temperature will cause the pressure in the pipe to increase. In the case of a leak, as the gas detection value increases, the tank pressure will slowly drop. Because of the correlation of these data, the change process of the eigenvalues ​​and eigenvectors of the data matrix in the data window will indicate different health states. In the characteristic model training module of the multi-dimensional joint monitoring algorithm, the characteristic models of the system in various health states (such as valve opening, leakage, high temperature, overpressure, etc.) are saved. As the window slides, the change process of the eigenvalue of the window data matrix is ​​correlated with the characteristic models of different health states, and the likelihood ratio between the real-time data model and various health state models can be obtained.

[0051] In order to overcome the bias caused by sensor noise data, in the real-time monitoring state judgment module of the multi-dimensional joint monitoring algorithm, the likelihood ratio value is accumulated for a length of L samples before the state judgment is made. As the data window slides, the result of the judgment output forms a monitoring state information data stream.

[0052] In the computing and storage module of the artificial intelligence on-site computing platform, the obtained monitoring status information data stream is stored in real time, and is simultaneously uploaded to the monitoring system platform through the mobile network access module of the artificial intelligence on-site computing platform.

[0053] The computing and storage module of the artificial intelligence on-site computing platform performs correlation statistical calculations on a large amount of sensor collected data, and the amount of status information data formed is less than one percent of the original sensor data, thereby reducing the demand for mobile communication network quality and bandwidth, and greatly improving the real-time and reliability of the monitoring method proposed in the present invention.

[0054] Step S4, the mobile network access module establishes a network data transmission channel with the monitoring system platform according to the wide area network access requirements, encodes the monitoring status information data stream obtained in step S3 to form transmission data, and sends the transmission data to the monitoring system platform through the established transmission channel.

[0055] This module encodes the monitoring status data, improving the reliability of data transmission over the wide area network and the security of information.

[0056] The above describes in detail the preferred implementation of this patent, but this patent is not limited to the above implementation. Within the scope of knowledge possessed by ordinary technicians in this field, this patent can also be implemented in other specific forms without departing from the purpose or basic characteristics of this patent.

Claims

1. A multi-dimensional pressure vessel health monitoring system based on artificial intelligence on-site calculation, characterized in that: It includes an artificial intelligence on-site computing platform, which is mainly composed of an on-site Internet of Things interface, a sensor data access module, a computing and storage module, and a mobile network access module. The computing and storage module includes a multi-dimensional joint monitoring module; wherein: The on-site IoT interface provides corresponding interfaces according to the communication modes of the N types of pressure vessel sensors to be connected, and automatically obtains the parallel real-time data collected by the sensors; The sensor data access module samples and normalizes the acquired parallel real-time data according to the requirements of the multi-dimensional joint monitoring module, and then uniformly encodes it; after uniform encoding, a continuous N-dimensional data stream is formed; through the sensor data access module, various types of sensor data are mapped into a unified multi-dimensional data space, which has the basis for correlation judgment of various types of sensor values; The computing and storage module stores the N-dimensional data stream in real time, and uses the multi-dimensional joint monitoring module to perform model processing on the N-dimensional data stream to form a monitoring status information data stream, and stores the obtained monitoring status information data stream in real time and sends it to the mobile network access module; The mobile network access module establishes a network data transmission channel according to the requirements of different types of wide area network access, encodes the monitoring status information data stream sent by the calculation and storage module to form transmission data, and sends the transmission data to the monitoring system platform through the established transmission channel; The multi-dimensional joint monitoring module includes a data window control module, a data model building module, a feature model training module, a real-time model likelihood ratio calculation module, and a real-time monitoring state judgment module, wherein: The data window control module controls the sampling window to slide at a set step length in the received N-dimensional data stream according to the sampling rate and the sampling window length L, and submits the N×L-dimensional data matrix in each sliding window to the data model building module for processing; The data model building module forms a real-time window data model by calculating the data space feature vector according to the N×L dimensional data matrix submitted by the data window control module, and submits the real-time window data model to the real-time model likelihood ratio calculation module; A real-time model likelihood ratio calculation module calculates the model space likelihood ratio of the received real-time window data model and the feature model, and simultaneously submits the likelihood ratio to the real-time monitoring state judgment module and the feature model training module; the feature model represents the local feature model of the system in a healthy state saved in the feature model training module; A real-time monitoring state judgment module, which uses the judgment criteria set by the multi-dimensional joint monitoring module to judge the likelihood ratio of the received current window data model, generates a current monitoring state data sequence of length N, and sends the monitoring state data sequence to the mobile network access module; The feature model training module uses the feature model initialized and trained by the multi-dimensional joint monitoring module as the starting feature model, and corrects the local feature model according to the spatial distribution of the feature model when receiving the likelihood ratio of each window data model; The data normalization includes time correction and sampling rate unification; The sampling window L of the data window control module is ≥ N. The selection of the specific sampling window length depends on the system's balance between monitoring accuracy and computing platform computing power.

2. The multi-dimensional pressure vessel health monitoring method based on the multi-dimensional pressure vessel health monitoring system based on artificial intelligence on-site calculation according to claim 1 is characterized in that: Here are the steps: Step 1: Install N sensors at the corresponding positions of the pressure vessel as required, and connect them to the on-site IoT interface to complete the unified adaptation of various access forms and electrical interfaces; Step 2: The sensor data access module interpolates the data of other sensors according to the shortest sampling period of each sensor, so that all sensor data are distributed between 0 and 1, and a unified sampling period of all sensor data is achieved, thus obtaining N parallel data streams. The interpolation is performed by linear interpolation or digital filtering interpolation; Then, through the sensor data access module of the artificial intelligence field computing platform, various types of sensor data are mapped into a unified multi-dimensional data space, thereby providing a basis for correlation judgment of various types of sensor values; Step 3, the computing and storage module first stores the N parallel data streams in real time, and at the same time uses the data window control module of the multi-dimensional joint monitoring module to select the sampling length of the data window L, and performs sliding window processing on the continuous parallel data streams to obtain an L×N sampling data matrix; The window sliding step size is 1, that is, the data matrix is ​​processed once for each sliding sampling point; as the window slides, the change process of the window data matrix eigenvalue is correlated with the characteristic model of different health states, and the likelihood ratio between the real-time data model and various health state models can be obtained; the result of the judgment output forms the monitoring state information data stream; Step 4, the mobile network access module establishes a network data transmission channel with the monitoring system platform according to the wide area network access requirements, encodes the monitoring status information data stream obtained in step 3 to form transmission data, and sends the transmission data to the monitoring system platform through the established transmission channel.

Images