Large-scale ship vibration isolation system, edge computing system and method

Abstract

The present invention discloses a large-scale vibration isolation system for ships, an edge computing system and a method. An edge node and an air pressure control module are configured for each airbag vibration isolation module. Each edge node is equipped with a displacement sensing module, an accelerometer module and an air pressure sensing module; the edge node inflates and deflates the airbag by controlling the air pressure control module. The edge node, the control node and the training node are connected to the switch using cables and exchange data through a local area network. The present invention utilizes the edge computing system as well as offline autonomous characteristics and data mining technology to propose an edge computing system for a large-scale vibration isolation system for ships. The edge computing system for a large-scale vibration isolation system for ships of the present invention, which has data mining, offline autonomy and load balancing capabilities, can better meet the functional requirements of a large-scale vibration isolation system.

Description

Technical Field

[0001] The present invention belongs to the technical field of ship safety control, and relates to a ship vibration isolation system, an edge computing system and a method, and specifically to a large-scale ship vibration isolation system, an edge computing system and a method. Background Art

[0002] In the ship power system, vibration has a huge negative impact on the ship and the accompanying crew. Vibration isolation devices are one of the main technical means to reduce vibration and noise of ships. As the requirements for intelligence and reliability of large-scale vibration isolation devices are further improved, a large number of sensors are used in vibration isolation systems, resulting in a significant increase in control system data throughput. For example, the existing control system based on the CAN bus network is difficult to carry such a large amount of data, resulting in data processing timeouts. In addition, a large amount of data has not been effectively used, but this information can provide an important reference for analyzing and predicting equipment status, fault warning and troubleshooting.

[0003] Under the background of ship modernization and intelligent development, the existing vibration isolation system has the characteristics of a large number of monitoring nodes, large computing load of the main control node, large data throughput and high network synchronization performance requirements, which limits the improvement and expansion of the control performance of the existing vibration isolation system.

[0004] At present, most of the methods or algorithms for improving the performance of vibration isolation systems are to actively control the current vibration isolation system by establishing a dynamic model. However, traditional data analysis methods such as establishing a dynamic model mostly sample or filter metadata and establish complex algorithms to obtain as much information as possible from limited samples. This method has problems such as long development cycle, difficult parameter tuning, and inaccurate model.

[0005] The ship vibration isolation system is developing towards large-scale and intelligent directions. The data acquisition, transmission and analysis architecture of the existing vibration isolation system severely limits the improvement of vibration isolation performance. Summary of the invention

[0006] In order to solve the above technical problems, the present invention utilizes the cloud-edge collaboration and offline autonomy characteristics and data mining technology of the edge computing system to propose a large-scale ship vibration isolation system, an edge computing system and a method. First, the edge computing system architecture is designed, and then the overall software and hardware architecture of the edge computing system is planned. The edge computing platform of the large-scale ship vibration isolation system with data mining, offline autonomy and load balancing capabilities of the present invention can better meet the functional requirements of large-scale vibration isolation systems.

[0007] The technical solution adopted by the large-scale vibration isolation system of the present invention is: a large-scale vibration isolation system of the ship, including a switch, a control node, a training node, a plurality of edge nodes, a plurality of airbag vibration isolation modules, a plurality of air pressure control modules, a plurality of displacement sensor modules, a plurality of accelerometer modules and a plurality of air pressure sensor modules, an air source, and an air source control valve;

[0008] The control node is connected to the switch and the training node respectively, and is used to control the entire system; the training node is used for deep learning model training, model management, and data set management; the plurality of edge nodes are arranged at preset positions in the ship; each edge node is electrically connected to the switch through a data transmission bus; each edge node is electrically connected to the airbag vibration isolation module and the air pressure control module in sequence, and the displacement sensing module, the acceleration module, and the air pressure sensing module are arranged in parallel between the edge node and the airbag vibration isolation module; the displacement sensing module is used to collect the displacement in the horizontal direction and the displacement in the vertical direction of the airbag vibration isolation module; the acceleration module is used to collect the acceleration in the horizontal direction and the acceleration in the vertical direction of the airbag vibration isolation module; the air pressure control module is used to inflate and deflate the airbag vibration isolation, and the air pressure sensing module is used to detect the air pressure in the airbag vibration isolation module in real time;

[0009] The gas source and the gas source control valve are electrically connected in sequence, and the gas source control valve is electrically connected to the plurality of gas pressure control modules one by one;

[0010] When the ship is running, the displacement sensor, accelerometer module and air pressure sensor are used to collect signals from the ship's power source, and the pressure value of the airbag vibration isolation module is adjusted according to the strength of the vibration to ensure that the vibration isolation system has a vibration reduction effect.

[0011] The technical solution adopted by the edge computing system of the present invention is: a large-scale vibration isolation edge computing system for ships, including a device layer, an edge computing layer, a network switching layer and an application layer;

[0012] The equipment layer is used for data acquisition, equipment management and I / O control, and is equipped with a large-scale ship vibration isolation system;

[0013] The edge computing layer is used for data processing, cluster management and edge computing; the data processing is to calculate and process the metadata obtained from the displacement sensing module, the acceleration module and the air pressure sensing module according to the loaded deep learning network, issue control commands to the execution device, and upload the metadata to the application layer for use as a data set after desensitization processing;

[0014] The application layer is used for vibration modeling, data visualization, fault warning, offline autonomy, data storage and load balancing control.

[0015] The technical solution adopted by the edge computing method of the present invention is: a large-scale ship vibration isolation edge computing method, which adopts a large-scale ship vibration isolation edge computing system; the equipment at the equipment layer accesses the cluster through the edge computing layer and exchanges data, the edge computing layer calculates the metadata, and uploads the processed data to the application layer through the network exchange layer, the application layer mines the received massive data through the deep learning network, obtains valuable information, performs fault warning, offline autonomy, data storage and load balancing control, and displays the obtained valuable information to the user;

[0016] The deep learning network includes a normalization processor, multiple feature extractors, a classifier and a softmax layer; the normalization processor is used to normalize the initial data sets of each sensor in the large-scale vibration isolation system of the ship to obtain the normalized vibration isolation system data x' i for:

[0017]

[0018] Among them, the sensor includes a displacement sensor module, an acceleration module and an air pressure sensor module. i is the sensor data obtained by the i-th edge node, x max is the maximum value in the sensor's initial historical data set, x min is the minimum value in the sensor's initial historical data set;

[0019] The multiple feature extractors are used to perform multiple feature extractions on the normalized vibration isolation system data to obtain key feature data;

[0020] The classifier is used to fuse multiple key feature data to obtain fused feature data;

[0021] The softmax layer is used to obtain the probability distribution of mechanical failure by fusing the feature data through the softmax layer to judge the mechanical failure.

[0022] Compared with the prior art, the beneficial effects of the present invention are: in view of the characteristics of the vibration isolation system, such as the large number of nodes, large amount of data, and high requirements for network synchronization performance, the edge computing is applied to the large-scale vibration isolation system of ships for the first time, and combined with data mining technology, an edge computing platform based on the large-scale vibration isolation system of ships is proposed, and the software and hardware hierarchical architecture of the platform is planned. The performance of the vibration isolation system is improved, the function of the vibration isolation system is expanded, and its vibration isolation effect is more superior. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a schematic diagram of a large-scale vibration isolation system for ships according to an embodiment of the present invention;

[0024] Figure 2This is a schematic diagram of a large-scale vibration isolation edge computing system for ships according to an embodiment of the present invention;

[0025] Figure 3 A schematic diagram of a cluster according to an embodiment of the present invention;

[0026] Figure 4 This is a load balancing control flow chart of an embodiment of the present invention;

[0027] Figure 5 The present invention provides a flowchart for obtaining a Nash equilibrium decision set according to an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to facilitate ordinary technicians in the field to understand and implement the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the implementation examples described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0029] At present, most of the methods or algorithms for improving the performance of vibration isolation systems are to actively control the current vibration isolation system by establishing a dynamic model. However, traditional data analysis methods such as establishing a dynamic model mostly sample or filter metadata and establish complex algorithms to obtain as much information as possible from limited samples. This method has problems such as long development cycle, difficult parameter tuning, and inaccurate model.

[0030] Different from traditional data analysis methods, the big data mining method of the present invention does not deliberately pursue the complexity and accuracy of the algorithm, but conducts a comprehensive analysis of all data sets. In addition, the four characteristics of the ship vibration isolation system: large volume, multiple modes, fast generation, huge value but low density belong to the category of big data. Through deep learning algorithms, more complex models can be constructed using massive data sets to effectively represent data.

[0031] Please see Figure 1 , a large-scale ship vibration isolation system provided in this embodiment includes a switch 1, a control node 2, a training node 3, a plurality of edge nodes 4, a plurality of airbag vibration isolation modules 5, a plurality of air pressure control modules 6, a plurality of displacement sensor modules 7, a plurality of acceleration modules 8 and a plurality of air pressure sensor modules 9, an air source 10, and an air source control valve 11;

[0032] The control node 2 is connected to the switch 1 and the training node 3 respectively, and is used to control the entire system; the training node 3 is used for deep learning model training, model management, and data set management; a number of edge nodes 4 are arranged at preset positions in the ship; each edge node 4 is electrically connected to the switch 1 through a data transmission bus; each edge node 4 is electrically connected to the airbag vibration isolation module 5 and the air pressure control module 6 in sequence, and a displacement sensing module 7, an acceleration module 8, and an air pressure sensing module 9 are arranged in parallel between the edge node 4 and the airbag vibration isolation module 5; the displacement sensing module 7 is used to collect the horizontal displacement and vertical displacement of the airbag vibration isolation module 5; the accelerometer module 8 is used to collect the horizontal acceleration and vertical acceleration of the airbag vibration isolation module 5; the air pressure control module 6 is used to inflate and deflate the airbag vibration isolation 5, and the air pressure sensing module 9 is used to detect the air pressure in the airbag vibration isolation module 5 in real time;

[0033] The gas source 10 and the gas source control valve 11 are electrically connected in sequence, and the gas source control valve 11 is electrically connected to a plurality of gas pressure control modules 6 one by one;

[0034] When the ship is running, the vibration signal of the ship's power source is collected through the accelerometer module 8, and the pressure value of the airbag vibration isolation module 5 is adjusted according to the strength of the vibration to ensure that the vibration isolation system has a vibration reduction effect.

[0035] Please see Figure 2 , a large-scale ship vibration isolation edge computing system provided in this embodiment includes a device layer, an edge computing layer, a network switching layer and an application layer;

[0036] The device layer of this embodiment is used for data acquisition, device management and I / O control, and is equipped with a large-scale ship vibration isolation system;

[0037] The edge computing layer of this embodiment is used for data processing, cluster management and edge computing; data processing, the metadata obtained from the displacement sensor module 7, the accelerometer module 8 and the air pressure sensor module 9 are calculated and processed according to the loaded deep learning network, and control commands are issued to the execution device, and the metadata is desensitized and uploaded to the application layer as a data set;

[0038] The edge computing layer of this embodiment is the core component of the edge computing system of the large-scale ship vibration isolation system. It relies on the edge computing system to sink the computing power to the vicinity of the data producer to ensure the real-time and security of data processing. Its main functions are:

[0039] (1) Cluster management of vibration isolation system nodes. A large vibration isolation system may have hundreds of nodes, and the hardware platforms of these hundreds of nodes may not be uniform, such as industrial computers, Raspberry Pi, etc. How to effectively manage the cluster is a key factor in whether the vibration isolation system can work as a whole.

[0040] (2) Data processing: The metadata obtained from the end side is calculated and processed according to the loaded deep learning model, and control commands are issued to the end side execution device. At the same time, the metadata is anonymized and uploaded to the application layer as a data set.

[0041] The application layer of this embodiment is used for vibration modeling, data visualization, fault warning, offline autonomy, data storage and load balancing control.

[0042] (1) Vibration modeling. With the help of deep learning algorithms, data mining is performed on the data sets uploaded by the edge computing layer to obtain more valuable hidden information. The data sets can also be used to train neural networks to establish vibration models, and the models can be sent to edge nodes for data prediction.

[0043] (2) Fault warning. The deep learning model established by data mining can realize functions such as fault warning and fault location by analyzing the data at the device layer, providing important reference for fault prevention and rapid recovery, and greatly improving the stability of the platform.

[0044] (3) Load balancing: When a node crashes or the computing load is too high, causing the CPU usage to reach the set threshold, part of the computing task will be offloaded to other suitable nodes.

[0045] This embodiment provides a large-scale ship vibration isolation edge computing method, which adopts a large-scale ship vibration isolation edge computing system; the characteristics are: the equipment at the equipment layer accesses the cluster through the edge computing layer and exchanges data, the edge computing layer calculates the metadata, and uploads the processed data to the application layer through the network exchange layer, the application layer mines the received massive data through the deep learning network, obtains valuable information, performs fault warning, offline autonomy, data storage and load balancing control, and displays the obtained valuable information to the user through the application layer;

[0046] The deep learning network of this embodiment includes a normalization processor, multiple feature extractors, a classifier and a softmax layer; the normalization processor is used to normalize the initial data sets of each sensor in the large-scale vibration isolation system of the ship to obtain the normalized vibration isolation system data x' i for:

[0047]

[0048] The sensor includes a displacement sensor module 7, an acceleration sensor module 8 and an air pressure sensor module 9. i is the sensor data obtained by the i-th edge node, x max is the maximum value in the sensor's initial historical data set, x min is the minimum value in the sensor's initial historical data set;

[0049] Multiple feature extractors, used for performing multiple feature extractions on normalized vibration isolation system data to obtain key feature data;

[0050] The classifier is used to fuse multiple key feature data to obtain fused feature data;

[0051] The softmax layer is used to obtain the probability distribution of mechanical failure by fusing feature data through the softmax layer to determine mechanical failure.

[0052] The devices in the device layer of this embodiment access the cluster through the edge computing layer and exchange data. Figure 3 , the cluster is composed of control node 2, training node 3 and all edge nodes 4.

[0053] The offline autonomy of this embodiment means that when an edge node is disconnected from the network and cannot exchange data with the control node, the edge node first tries to reconnect. If it fails, it continues to run according to the current work task and saves the sensor data locally after desensitizing it. The control node sends a warning to the administrator after detecting that the edge node is offline. After the edge node comes back online, it first synchronizes its own status with the control node, and then uploads the saved sensor data to the training node 3 to complete the data synchronization.

[0054] Please see Figure 4 The load balancing control of this embodiment can guarantee the distributed computing requirements of the large vibration isolation system of the ship. When the load of the main computing node and the training node is too large, the load balancing algorithm can be used to unload the distributed computing tasks. Each computing task will be scheduled to the node with idle resources for joint calculation, thereby reducing the overall computing time.

[0055] In order to ensure the real-time performance of the calculation task of large-scale vibration isolation of ships, the main optimization target of this load balancing algorithm is delay. It is based on single-step optimal decision control and the "greedy principle". Each calculation task searches for the node that can minimize its own calculation delay in turn according to the current load of each node to unload the calculation. Through iteration, when each task cannot find a node that can reduce its own calculation time, it is considered to have achieved load balancing, and the overall calculation time will also be smaller; the specific implementation includes the following steps:

[0056] (1) Select the initial strategy, that is, choose local computing instead of offloading;

[0057] (2) Receive resource information from all nodes;

[0058] (3) When making the first round of decisions or receiving the decision information of the previous round, the utility values ​​of all nodes that can be offloaded are calculated;

[0059] (4) Select the decision with the highest utility value;

[0060] (5) If the decision set of all tasks in this round is the same as the decision set of the previous round, load balancing is achieved; otherwise, the process returns to step (1).

[0061] Please see Figure 5 The decision set of this embodiment is a Nash equilibrium decision set, and its acquisition process is to make a single-step optimal decision in turn through each node, and cyclically update the decision to finally reach a Nash equilibrium decision set.

[0062] Nash equilibrium decision set, the utility calculation formula is as follows:

[0063]

[0064] U n (m,S n ,H -n ) indicates that task n is in decision S n The delay after each computing task is offloaded to the target idle node m is Rather than choosing a delay for uninstallation The sum of the differences;

[0065] Each node adopts the single-step optimal decision function as shown below:

[0066]

[0067] Where m * is the optimal node that can be selected for this task, H -n Represents the decision information for each task except this task.

[0068] The deep learning network of this embodiment is a trained deep learning network; the loss function used in the training is the cross entropy loss function;

[0069] The cross entropy loss function is:

[0070]

[0071] Where M represents the number of categories; y ia represents the sign function, if the true category of sample i is equal to a, it takes 1, otherwise it takes 0; p ia represents the probability that training sample i belongs to category a, and N is the total number of training samples.

[0072] The present invention uses edge computing system, offline autonomous characteristics and data mining technology to propose an edge computing system for large-scale vibration isolation system of ships. First, the system architecture is designed, and then the overall software and hardware architecture of the system is planned. The edge computing system for large-scale vibration isolation system of ships with data mining, offline autonomy and load balancing capabilities of the present invention can better meet the functional requirements of large-scale vibration isolation systems.

[0073] It should be understood that the above description of the preferred embodiment is relatively detailed and cannot be regarded as limiting the scope of patent protection of the present invention. Under the enlightenment of the present invention, ordinary technicians in this field can also make substitutions or modifications without departing from the scope of protection of the claims of the present invention, which all fall within the scope of protection of the present invention. The scope of protection requested for the present invention shall be based on the attached claims.

Claims

1. A large-scale vibration isolation edge computing method for ships, applied to a large-scale vibration isolation edge computing system for ships; characterized in that: The large-scale vibration isolation edge computing system for ships includes a device layer, an edge computing layer, a network switching layer and an application layer; the devices in the device layer access the cluster and exchange data through the edge computing layer, the edge computing layer calculates the metadata, and uploads the processed data to the application layer through the network switching layer, the application layer mines the received massive data through the deep learning network, obtains valuable information, performs fault warning, offline autonomy, data storage and load balancing control, and displays the obtained valuable information to the user; The deep learning network includes a normalization processor, multiple feature extractors, a classifier and a softmax layer; the normalization processor is used to normalize the initial data sets of each sensor in the large-scale vibration isolation system of the ship to obtain the normalized vibration isolation system data x i 'for: The sensor includes a displacement sensor module (7), an accelerometer module (8) and an air pressure sensor module (9). i is the sensor data obtained by the i-th edge node, x max is the maximum value in the sensor's initial historical data set, x min is the minimum value in the sensor's initial historical data set; The multiple feature extractors are used to perform multiple feature extractions on the normalized vibration isolation system data to obtain key feature data; The classifier is used to fuse multiple key feature data to obtain fused feature data; The softmax layer is used to obtain the probability distribution of mechanical failure by fusing the feature data through the softmax layer to determine the mechanical failure; The load balancing control is used to ensure the distributed computing requirements of the large vibration isolation system of the ship. When the load of the edge node (4) is too large, the distributed computing tasks are unloaded through the load balancing control, and each computing task will be dispatched to the edge node (4) with idle resources for joint computing, thereby reducing the overall computing time. In order to ensure the real-time performance of the calculation task of large-scale vibration isolation of ships, the main optimization target of this load balancing control is delay. It is based on single-step optimal decision control and the "greedy principle". Each calculation task searches for the node that can minimize its own calculation delay in turn according to the current load situation of each node to unload the calculation; through iteration, when each task cannot find a node that can reduce its own calculation time, it is considered to have achieved load balancing, and the overall calculation time will also be smaller at this time; the specific implementation includes the following steps: (1) Select the initial strategy, that is, choose local computing instead of offloading; (2) Receive resource information of all nodes; (3) When making the first round of decisions or receiving the decision information of the previous round, the utility values ​​of all nodes that can be offloaded are calculated; (4) Choose the decision with the highest utility value; (5) If the decision set of all tasks in this round is the same as the decision set of the previous round, load balancing is achieved; otherwise, the process returns to step (1).

2. The large-scale vibration isolation edge calculation method for ships according to claim 1 is characterized in that: The devices at the device layer access the cluster through the edge computing layer and exchange data. The cluster is composed of a control node (2), a training node (3) and all edge nodes (4).

3. The large-scale vibration isolation edge calculation method for ships according to claim 1 is characterized in that: In the offline autonomy, when an edge node network is disconnected and cannot exchange data with the control node, the edge node first attempts to reconnect. If it fails, it continues to run according to the current work task and saves the sensor data to the local database after desensitization. After the control node detects that the edge node is offline, it issues a warning to the administrator. After the edge node comes back online, it first synchronizes its own status with the control node, and then uploads the saved sensor data to the training node (3) to complete data synchronization.

4. The large-scale vibration isolation edge calculation method for ships according to claim 3 is characterized in that: The decision set is a Nash equilibrium decision set, and the acquisition process thereof is to make single-step optimal decisions in turn through each node, and cyclically update the decisions to finally reach the Nash equilibrium decision set.

5. The large-scale vibration isolation edge calculation method for ships according to claim 4 is characterized in that: The utility calculation formula of the Nash equilibrium decision set is as follows: U n (m,S n ,H -n ) means that task n is in decision S n The delay after each computing task is offloaded to the target idle node m is Instead of not choosing the uninstall delay The sum of the differences; Each node adopts the single-step optimal decision function as shown below: Where m * is the optimal node that can be selected for this task, H -n Represents the decision information for each task except this task.

6. The large-scale vibration isolation edge computing method for ships according to any one of claims 1 to 5, characterized in that: The deep learning network is a trained deep learning network; the loss function used in the training is a cross entropy loss function; The cross entropy loss function is: Where M represents the number of categories; y ia represents the sign function, if the true category of sample i is equal to a, it takes 1, otherwise it takes 0; p ia represents the probability that training sample i belongs to category a, and N is the total number of training samples.

7. The large-scale vibration isolation edge computing method for ships according to any one of claims 1 to 6, characterized in that: The large-scale vibration isolation edge computing system for ships includes a device layer, an edge computing layer, a network switching layer and an application layer; The equipment layer is used for data acquisition, equipment management and I / O control, and is equipped with a large-scale ship vibration isolation system; The edge computing layer is used for data processing, cluster management and edge computing; the data processing is to calculate and process the metadata obtained from the displacement sensor module (7), the accelerometer module (8) and the air pressure sensor module (9) according to the loaded deep learning network, issue control commands to the execution device, and upload the metadata to the application layer for use as a data set after desensitization processing; The application layer is used for vibration modeling, data visualization, fault warning, offline autonomy, data storage and load balancing control.

8. The large-scale vibration isolation edge calculation method for ships according to claim 7 is characterized in that: The large-scale ship vibration isolation system comprises a switch (1), a control node (2), a training node (3), a plurality of edge nodes (4), a plurality of airbag vibration isolation modules (5), a plurality of air pressure control modules (6), a plurality of displacement sensor modules (7), a plurality of accelerometer modules (8) and a plurality of air pressure sensor modules (9), an air source (10), and an air source control valve (11); The control node (2) is respectively connected to the switch (1) and the training node (3) for controlling the entire system; the training node (3) is used for deep learning model training, model management, and data set management; the plurality of edge nodes (4) are arranged at preset positions in the ship; each edge node (4) is electrically connected to the switch (1) via a data transmission bus; each edge node (4) is electrically connected to an airbag vibration isolation module (5) and an air pressure control module (6) in sequence, and the edge node (4) and the airbag vibration isolation module (5) are arranged in parallel. The invention comprises a displacement sensing module (7), an accelerometer module (8) and an air pressure sensing module (9); the displacement sensing module (7) is used to collect the displacement in the horizontal direction and the displacement in the vertical direction of the airbag vibration isolation module (5); the accelerometer module (8) is used to collect the acceleration in the horizontal direction and the acceleration in the vertical direction of the airbag vibration isolation module (5); the air pressure control module (6) is used to inflate and deflate the airbag vibration isolation module (5); and the air pressure sensing module (9) is used to detect the air pressure in the airbag vibration isolation module (5) in real time; The gas source (10) and the gas source control valve (11) are electrically connected in sequence, and the gas source control valve (11) is electrically connected one by one to the plurality of gas pressure control modules (6); When the ship is running, the displacement sensor module (7), the accelerometer module (8) and the air pressure sensor module (9) collect signals from the ship's power source, and the pressure value of the airbag vibration isolation module (5) is adjusted according to the strength of the vibration, thereby ensuring that the vibration isolation system has a vibration reduction effect.

Images