A Ship Power Data Processing System, Method and Medium

Abstract

The present invention relates to the technical field of ships, and provides a ship power data processing system, method and medium. The system includes: an analog-to-digital conversion module that converts the collected ship engine room analog signals into digital signals by connecting sensors; a data caching module and a gigabit Ethernet module based on ZYNQ, which are used to cache high-speed digital signals to prevent signal distortion and communicate with the upper computer to send the digital signals completely to the intelligent management module; an intelligent management module that is respectively connected to the database and the signal acquisition module, realizes remote monitoring of the engine room, compares with the fault data in the database to realize the fault diagnosis function, and gives relevant operation and maintenance suggestions according to the diagnosis results. This system has greatly improved the intelligent level of the ship engine room, and has improved the intelligence and safety of ship navigation compared with the maintenance method based on manual experience.

Description

Technical Field

[0001] The present invention relates to the technical field of ships, and particularly relates to a ship power data processing system, method and medium. Background Art

[0002] The ship engine room is the core and center of green ship power. Whether the engine room equipment can operate normally directly affects the operation state of the ship. Traditional engine room management relies on manual operation and regular maintenance. Therefore, ship intelligence is an important trend for the future development of ships. As the most important part of ship power intelligence, the intelligent engine room uses advanced technologies and automated systems to improve the efficiency, reliability, safety and maintainability of the engine room, mainly involving technologies such as sensors, automated data analysis and remote monitoring, enabling the engine room to monitor, adjust and diagnose faults in real time and automatically, greatly improving the intelligent level of ship operation.

[0003] In the current related prior arts, most relevant staff rely on the operation state and experience of engine room equipment to alarm and repair ships, which cannot meet the requirements of intelligent and safe ship power management. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to make up for the lack of experience in designing a domestic self-developed high-speed, high-precision and high-real-time engine room data acquisition card in the prior art, and the low intelligent level of ship power management caused by the staff relying on experience for ship operation and maintenance. Therefore, a ship power data processing system, method and medium are provided.

[0005] To solve the above technical problem, the present invention provides a ship power data processing system based on ZYNQ-DSP as the data acquisition card and processing core. The management object of the ship power data processing system is the equipment to be monitored in the engine room. The ship power data processing system includes: a database, which stores the signal data into the database when detecting sudden situations such as faults of the equipment to be monitored in the engine room, and periodically stores the daily data and establishes a signal comparison module, outputs the diagnosis result through the traditional method of threshold judgment for other engine room signals, and compares it with the intelligent fault diagnosis algorithm based on vibration signals proposed by the system to reflect the algorithm advantages; a hardware acquisition unit, which is suitable for collecting and processing the data and signals generated after the operation of the engine room equipment, communicating with the intelligent management unit, and transmitting the collected data of the engine room equipment to the upper computer; an intelligent management unit, which is suitable for receiving the real-time operation state signals of the engine room equipment sent by the hardware acquisition unit, processing them to obtain the corresponding physical state information, and at the same time using the intelligent diagnosis algorithm to process the vibration signals and comparing them with the corresponding fault data in the database to judge whether there is equipment failure in the engine room, making corresponding evaluations, and giving corresponding intelligent operation and maintenance suggestions to relevant personnel.

[0006] Optionally, the hardware acquisition unit includes: a data acquisition module, which is suitable for collecting analog signals of the operating state of the equipment to be monitored in the engine room through sensors and converting them into digital signals and sending them to the data processing module; a core board module, which is suitable for controlling the data acquisition module to read sensor signals by sending control signals, and processing the electrical signals after receiving the electrical signals converted by the data acquisition module; a data communication module, which is controlled by the core board module, and uploads the processed engine room signals to the intelligent management unit for display;

[0007] Optionally, the ship power data processing system further includes a sensor module. The sensor module is suitable for collecting analog signal outputs of the equipment to be monitored in the engine room. The acquisition module includes: a signal conditioning module, which is located between the sensor and the A / D analog-to-digital conversion module. The signal conditioning module is suitable for receiving the analog signals output by the sensor component and performing operations such as amplification, filtering, conversion, and isolation on the analog signals of the equipment to be monitored in the engine room; an A / D analog-to-digital conversion module, which monitors the operating state of the equipment to be monitored in the engine room through sensors, converts the collected vibration analog signals into digital signals and communicates them to the ZYNQ module, and converts signals other than vibration into digital signals and communicates them to the DSP module;

[0008] Optionally, the ship power data processing system further includes a core board module. The core board module includes: a ZYNQ module, which communicates with the A / D analog-to-digital conversion module to receive vibration signals, stores the high-frequency vibration signals in the data cache module DDR3 through the AXI4 bus and FIFO (first in first out), and reads the cached data and transmits the vibration signals to the intelligent management unit through the UDP protocol module; a DSP module, which receives the vibration signals transmitted by the ZYNQ module and collects the non-vibration signals of the equipment to be monitored in the engine room. After writing the signals into the signal processing module, it transmits the processed signals to the intelligent management unit through the RS232 communication module; the vibration signals collected by the ZYNQ module are transmitted to the DSP module through a master-slave multi-channel SPI communication method.

[0009] Optionally, the core board module further includes a signal processing module, which is suitable for the DSP module to process the vibration signals transmitted by the ZYNQ module and the non-vibration signals collected by the equipment to be monitored in the engine room, obtain the one-dimensional time series characteristics of the vibration signals and the physical state information of other non-vibration signals, and then send them to the intelligent management unit for fault diagnosis through intelligent algorithms.

[0010] Optionally, the intelligent management unit includes: an online monitoring module, which is communicatively connected to the data communication module. The online monitoring module receives the signals uploaded by the ZYNQ module and the DSP module, and after performing relevant processing, obtains corresponding physical information, thereby realizing the online monitoring of the operating status of the equipment to be monitored in the engine room; a data storage module, which is communicatively connected to the database and the core board module. The data storage module is adapted to store the signal data into the database when detecting sudden situations such as failures of the equipment to be monitored in the engine room, and perform periodic storage of daily data.

[0011] Optionally, the intelligent management unit further includes: a fault diagnosis module, which is communicatively connected to the database and the core board module, analyzes the signals fed back by the equipment to be monitored in the engine room through an intelligent diagnosis algorithm model, and diagnoses whether a failure occurs and the degree of the failure; an intelligent operation and maintenance module, which is adapted to, after the fault diagnosis module determines whether a failure occurs and the degree of the failure according to the operation data of the equipment to be monitored in the engine room, evaluate according to the algorithm output result and give operation and maintenance suggestions to the staff.

[0012] Among them, the intelligent fault diagnosis algorithm is a fault diagnosis algorithm based on Markov Transition Field (MTF for short) and Reparameterized Local Kernel Networks (RepLKNet for short), and its steps are as Figure 3 shown.

[0013] First, the Markov Transition Field MTF converts the one-dimensional time series signal into a two-dimensional time-frequency image feature. It is assumed that the transition matrix M of the system with N discrete states is an N×N matrix, where each element represents the transition probability from state to state . If this matrix is defined by the Q quantile, then its dimension is Q×Q.

[0014] Specifically, the Markov Transition Field MTF assumes there is a time series as follows:

[0015]

[0016] Among them, each is the state at each moment, is the length of the sequence.

[0017] Then the time series is divided into N quantile segments, labeled 1, 2,..., N, and then The data in each is corrected to the corresponding quantile segment markers. Subsequently, a Markov state transition matrix is constructed , and the element in the row and column represents the probability that the quantile segment

[0018]

[0019] transfers to

[0020]

[0021] Based on this state transition matrix, MTF is given:

[0022] The value of the Markov transformation field MTF is directly imaged into a grayscale image or mapped into a pseudo-color image, that is, the transformation from a one-dimensional time series signal to an image is realized, and two-dimensional time-frequency features are obtained.

[0023] Then, the one-dimensional time series features and the two-dimensional time-frequency features obtained by the Markov transformation field MTF transformation are combined, and the features are respectively input into the neural network RepLKNet based on large convolutional kernels for training, and the computational efficiency and classification performance are improved through local convolutional kernels and reparameterization techniques.

[0024] Among them is the one-dimensional time series feature, with a shape of ( , ), where is the number of samples, is the feature length of each sample; is the two-dimensional image feature, with a shape of ( , , , ), where is the number of samples, , are the height and width of the image respectively, is the number of channels of the image.

[0025] These two types of features (time series features and image features) are concatenated to generate a composite feature representation for classification.

[0026] One-dimensional time series feature Local features can be extracted through some convolution operations or directly converted into a lower-dimensional vector through a fully connected layer. Suppose it is converted into a feature vector with dimensions:

[0027]

[0028] where , is the convolution operation, is the convolution kernel.

[0029] Two-dimensional image features After being processed by the local convolution operation in RepLKNet in the convolutional neural network, a tensor with a shape of ( , , , ) is obtained. To concatenate with one-dimensional features, it is usually necessary to perform pooling or convolution on the image features to adjust their dimensions. Suppose the processed image features are , and its shape is ;

[0030]

[0031] where , is the pooling operation.

[0032] The processed one-dimensional time series features and the two-dimensional image features are concatenated in the feature dimension to obtain a new feature representation :

[0033]

[0034] The concatenated features can be passed into a local convolutional layer to extract finer-grained local features. The concatenated features are used as the input to the convolutional layer to further learn local patterns.

[0035] The neural network RepLKNet based on large convolution kernels introduces large convolution kernels, which can more effectively capture long-range context information. Among them, the large convolution kernels are reparameterized, usually by decomposing larger convolution kernels into smaller convolution operations (such as 1x1 convolutions and other local convolutions) to reduce computational overhead. The reparameterized convolution can be represented in a form similar to the following:

[0036]

[0037] where It is the convolution kernel after reparameterization. The core is to decompose complex convolution operations into a set of simple and more efficient operations.

[0038] The neural network RepLKNet based on large convolution kernels uses local convolution kernels to replace traditional large convolution kernels. Suppose there is a local convolution kernel , the output of this local convolution can be expressed as:

[0039]

[0040] where is the local convolution kernel. Local convolution captures local features by restricting the size of the convolution kernel to a smaller local area, thus reducing the computational amount.

[0041] The neural network RepLKNet based on large convolution kernels extracts features at different levels through convolution operations at multiple scales. Usually, the network starts with smaller local convolutions and then aggregates these features through global convolution or pooling operations. Let the convolution operations at different scales be:

[0042]

[0043] Finally, the trained concatenated features are input into the fully connected layer for classification operations:

[0044]

[0045] where, is the final classification output, that is, the output fault category, is the weight matrix of the fully connected layer, is the bias term, The function is used to map the output to the category probability space.

[0046] During the training process, the neural network RepLKNet based on large convolution kernels adopts the technique of adaptive noise to enhance the robustness of local convolution. Noise is added to the input and its intensity is adjusted in each training iteration. The input after adding noise is expressed as:

[0047]

[0048] where, is the noise generated as needed, usually gradually decreasing during the training process. In this way, the neural network RepLKNet based on large convolution kernels can increase robustness and reduce mode aliasing when processing signals.

[0049] The loss function of the neural network RepLKNet based on large convolutional kernels is usually based on common classification loss functions, such as cross-entropy loss. In a classification task, the cross-entropy loss function is as follows:

[0050]

[0051] where is the true label, is the predicted probability output by the network, is the number of classes, and e is the natural logarithm. Using the natural logarithm as the base helps calculate the gradient during the optimization process and avoids unnecessary complexity.

[0052] Optionally, the database includes a set of vibration signal data samples and non-vibration signal normal operation and fault range data during normal operation and different faults of cabin equipment collected through device parameters, data simulation, and laboratory simulation experiments, and a life cycle degradation data sample set established according to the component degradation law of the equipment to be monitored in the cabin; at the same time, when a sudden situation such as a fault occurs in the equipment to be monitored in the cabin, its signal data is stored in the database, and the daily data is stored periodically.

[0053] This application also provides a method for processing ship power data. Using the ship power data processing system based on the ZYNQ module and the DSP module, the method for processing ship power data includes: collecting the operating state signals of the equipment to be monitored in the cabin except for vibration signals (such as voltage, temperature, pressure, etc.); collecting the vibration signals of equipment such as engines, propulsion shafts, generators, motors, ventilation units, and pumps in the equipment to be monitored in the cabin; analyzing and processing the operating state data signals, and realizing online monitoring of the equipment to be monitored in the cabin after obtaining the data processing results; comparing the signals other than vibration with the normal operating parameters of the equipment to obtain a conclusion on whether a fault has occurred, and putting the processed vibration signals into an intelligent diagnosis algorithm to obtain a fault diagnosis result; outputting corresponding maintenance suggestions according to the fault diagnosis result.

[0054] Optionally, the method for processing ship power data further includes: storing the signal data in the database when a sudden situation such as a fault occurs in the equipment to be monitored in the cabin, and storing the daily data periodically.

[0055] This application also provides a computer storage medium, a hardware device suitable for storing data and programs, and the program realizes the steps of the above method when executed.

[0056] The technical solution of the present invention has the following advantages:

[0057] 1. The ship power data processing system provided by the present invention designs a high-speed data acquisition and processing module with a ZYNQ-DSP architecture, which meets the requirements of intelligent cabin data acquisition. It can collect data on various signals of the operating status of cabin equipment, process the signal data, and then transmit the data to the intelligent management unit to achieve real-time monitoring of the operating status of cabin equipment. Through real-time status monitoring and intelligent fault diagnosis algorithms, it diagnoses the fault of the equipment vibration signal, gives intelligent operation and maintenance suggestions to relevant personnel, and improves the reliability and intelligent level of ship operation.

[0058] 2. The ship power data processing system provided by the present invention has vibration signals sourced from rotating mechanical equipment in the cabin such as engines, pumps, compressors, bearings, and couplings collected by the hardware acquisition unit. The vibration signals can effectively reflect the health status of such equipment. The faults or performance degradation of the equipment usually cause changes in vibration characteristics, such as frequency, amplitude, and waveform changes, which can effectively reflect the status of mechanical components. Compared with other diagnostic methods (such as temperature, oil analysis, etc.), vibration monitoring has the advantages of non-invasiveness and no need for shutdown. At the same time, a database is designed to achieve timely storage of fault data and periodic preservation of daily data, ensuring the normal operation of cabin equipment, improving the troubleshooting efficiency of faulty equipment, and setting up a signal comparison module. The signal comparison module is suitable for outputting diagnostic results of other cabin signals through traditional methods of threshold judgment and comparing them with the intelligent fault diagnosis algorithm based on vibration signals proposed by the system. After a fault occurs, the traditional threshold method only judges whether a mechanical equipment fault has occurred, and the output result is inaccurate when the fault is minor. The intelligent fault diagnosis algorithm of the system can accurately judge whether a fault has occurred, whether the fault is minor or serious, and can also judge which component of the equipment has a fault.

[0059] 3. The ship power data processing system provided by the present invention avoids the situation where staff still need to troubleshoot and repair manually and based on experience after a fault occurs through real-time perception monitoring and fault warning of the cabin environment and equipment operating status, improves the intelligent level and operation and maintenance management efficiency of the ship, reduces economic losses, improves the comprehensive energy efficiency of the ship, and makes ship navigation more economical. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0061] Figure 1It is a framework diagram of a ship power data processing system according to Embodiment 1 of the present invention;

[0062] Figure 2 It is a framework diagram of the hardware acquisition unit provided in Embodiment 1 of the present invention;

[0063] Figure 3 It is a logic flow chart of the ship power data processing method provided in Embodiment 1 of the present invention. Detailed implementation manners

[0064] Next, the technical solutions of the present invention will be clearly and completely described in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0065] In addition, the technical features involved in different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0066] Application overview

[0067] At present, the processing of ship power data is still in the initial development stage, and the safety guarantee system of intelligent ships is still being improved. How to improve the stability of the system, remotely monitor and fault diagnose the engine room equipment of green ships, etc. are all key problems to be solved urgently. For ship power data processing technology, the ship engine room is the core of this technology. To realize the online monitoring of the engine room operation status, the engine room signal acquisition system is in a core position. Especially in some fields such as transient signal detection, there is an increasing need for a set of self-developed data acquisition and processing modules that meet high speed, high precision, and high real-time performance to achieve the function indicators of engine room signal acquisition. In addition, with the continuous deepening of the concept of ship intelligence, the intelligent power management of green ships not only satisfies the online monitoring of engine room electromechanical equipment, but also needs to use the signal data collected by the high-speed data acquisition module and intelligent fault diagnosis algorithms to judge whether the equipment has failed, so as to realize ship intelligence and improve the reliability of ship operation.

[0068] In addition, according to the Shannon sampling theorem, to ensure the integrity of the sampled signal, the signal sampling frequency should be kept more than twice the frequency of the signal itself. And the engine room equipment such as the bearing vibration signal can reach thousands or even tens of thousands of hertz. The current engine room power management system cannot achieve the complete data acquisition and processing of high speed, high precision, and high real-time performance, and lacks an integrated solution. This application will propose a solution to the above technical defects.

[0069] Among them, for the solution to the problem, first, it is necessary to sort out the equipment parameters and characteristics in the ship's engine room. For the operating states of different mechanical and electrical equipment and different components, what monitoring schemes should be adopted and transmitted into the intelligent management unit. For data such as temperature, pressure, and voltage, alarms should be set with reference to the operating parameters of relevant equipment. For high-frequency vibration signals, signal processing and intelligent algorithms will be carried out by comparing the full-life data, so as to realize the fault diagnosis and health assessment of the equipment. Corresponding intelligent operation and maintenance methods will be proposed for the operating states and fault diagnosis results of different engine room equipment, avoiding inefficient maintenance methods that rely on manual labor and experience.

[0070] Regarding the solution to the above difficulties, its significance lies in that nowadays the ship industry is gradually developing towards the directions of "intelligentization" and "greenization", and a set of intelligent engine room management system based on a high-speed and high-precision data acquisition card is the core to realize ship intelligentization. To overcome the technical defects in the prior art that lack the experience of designing domestic self-developed high-speed, high-precision, and high-real-time engine room data acquisition cards, and that the staff rely on experience for ship operation and maintenance, resulting in low reliability and intelligentization of the ship. Through this system, the operating states of relevant equipment in the engine room can be monitored online, fault diagnosis can be realized through intelligent algorithms, operation and maintenance suggestions can be put forward, and data can be stored in a timely manner for subsequent analysis, assisting the crew in efficiently repairing the equipment, deepening the degree of ship intelligentization, and improving the reliability of ship navigation.

[0071] Embodiment 1

[0072] As Figure 1 shown, a ship power data processing system based on an extensible processing platform (ZYNQ) and a high-performance 32-bit floating-point operation processor (Digital Signal Processing, abbreviated as DSP). There are engine room equipment to be monitored in the ship's engine room. The ship power data processing system includes: a hardware acquisition unit, a database, and an intelligent management unit. The hardware acquisition unit is a designed high-speed acquisition card, suitable for collecting the analog quantities of the mechanical and electrical equipment in the engine room equipment to be monitored, processing the analog quantities into corresponding electrical signals and communicating with the intelligent management unit; the intelligent management unit is suitable for receiving the real-time operating state signals of the engine room equipment sent by the hardware acquisition unit, processing them to obtain the corresponding physical state information, and at the same time using intelligent diagnostic algorithms to process the vibration signals and compare them with the corresponding fault data in the database to judge whether there is a fault in the engine room equipment, and making corresponding evaluations, and giving corresponding intelligent operation and maintenance suggestions to relevant personnel; the database is suitable for storing the signal data into the database when sudden situations such as faults in the engine room equipment to be monitored are detected, and storing the daily data periodically.

[0073] The above-mentioned equipment to be monitored in the engine room includes a variety of intelligent marine engine room electromechanical equipment, and these equipment will be regarded as the equipment to be monitored in the engine room. The equipment to be monitored in the engine room includes engines, propulsion shaft systems, generators, motors, filters, fuel systems, pumps, cooling systems, starting and control air systems, ventilation units, lubricating oil systems, and fresh water booster devices. Different electromechanical equipment corresponds to different analog signals. Specifically, the corresponding analog signals respectively include engine temperature, pressure, and vibration signals, propulsion shaft system temperature and vibration signals, generator temperature, voltage, current, speed, and vibration signals, motor vibration and temperature signals, filter temperature and pressure signals, fuel system pressure signals, pump vibration, temperature, pressure, and flow signals, cooling system temperature and pressure signals, starting and control air system pressure signals, ventilation unit vibration, flow, and temperature signals, lubricating oil system temperature and pressure signals, and fresh water booster device pressure signals.

[0074] Among them, as Figure 2 shown, corresponding sensors are selected to read the analog signals of different equipment in the equipment to be monitored in the engine room. The engine is correspondingly equipped with temperature, pressure, and vibration sensors for obtaining the vibration signal of the cylinder head in the engine, the temperature and pressure signals of the cylinder liner and supercharger, and the temperature signal of the fuel injection valve; the propulsion shaft system is correspondingly equipped with temperature and vibration sensors, which can be used to obtain the temperature and vibration signals of the shaft, bearings, and transmission gears in the propulsion shaft system; the generator is correspondingly equipped with temperature, voltage, current, speed, and vibration sensors for obtaining the stator voltage, current, and temperature signals, rotor vibration signal, and speed signal in the generator; the motor is correspondingly equipped with vibration and temperature sensors for obtaining the motor vibration and temperature signals; temperature and pressure sensors are respectively arranged at the inlets and outlets of various filters to obtain the inlet and outlet temperature and pressure information for monitoring the heat exchange efficiency; pressure or differential pressure sensors are respectively arranged at the inlets and outlets of each equipment in the fuel system for obtaining the differential pressure to observe the fuel supply capacity and heat exchange capacity; the pumps include cooling pumps, fire pumps, main engine jacket water pumps, etc., and are correspondingly equipped with vibration, temperature, pressure, and flow sensors for obtaining vibration, temperature, pressure, and flow signals to monitor the cooling medium supply capacity or water supply and drainage capacity of each pump; the cooling system is correspondingly equipped with temperature and pressure sensors for monitoring the inlet and outlet pressure of the condensate pump and filter and the temperature of the heat exchanger to observe the cooling medium supply capacity and heat exchange performance; the starting and control air system is correspondingly equipped with a pressure sensor for monitoring the outlet pressure of the air storage tank and compressor and the lubricating oil pressure to monitor the equipment air supply capacity; the ventilation unit is correspondingly equipped with vibration, flow, and temperature sensors for monitoring the vibration, temperature, and pressure signals of the air supply and exhaust fans to ensure the engine room ventilation capacity; the lubricating oil system is correspondingly equipped with temperature and pressure or differential pressure sensors for measuring the pressure difference and temperature at the inlet and outlet of the equipment to monitor the lubricating oil supply capacity; the fresh water booster device is correspondingly equipped with a pressure sensor for measuring the outlet pressure of the storage tank and booster pump to ensure the water supply capacity.

[0075] The ship power data processing system uses a hardware acquisition unit to collect various analog signals of the equipment to be measured in the engine room, and after processing the analog signals, communicates them to the upper computer. The intelligent management unit of the upper computer processes the received data and realizes the online real-time monitoring of the operating status of the engine room equipment through visual operations. It uses intelligent algorithms to analyze the signals for fault diagnosis, outputs corresponding operation and maintenance suggestions according to the results, and periodically stores the engine room data and stores the data and nodes in case of faults for subsequent analysis and evaluation, which deepens the intelligent level of ship engine room management and improves the reliability of ship navigation.

[0076] The hardware acquisition unit is installed in the cabinet of the ship power data processing system, and the upper computer program of the intelligent management unit is deployed on the industrial panel. The panel is communicatively connected to the hardware acquisition unit through the serial port to USB and network cable methods.

[0077] Furthermore, the hardware acquisition unit includes: a data acquisition module, a core board module, and a data communication module. The data acquisition module is used to read different analog signals of the engine room equipment and convert them into electrical signals that the core board module can process through signal processing and analog-to-digital conversion. The core board module processes and caches the signals and then sends the signals to the intelligent management unit of the upper computer through the data communication module.

[0078] Among them, the data acquisition module includes: a signal conditioning module and an A / D analog-to-digital conversion module. The signal conditioning module is located between the sensor component and the A / D analog-to-digital conversion module. The signal conditioning module is suitable for performing operations such as amplification, filtering, conversion, and isolation on the analog signals of the engine room equipment to be monitored collected. The A / D analog-to-digital conversion module converts the vibration signals processed by the signal conditioning module into digital signals and communicates them to the ZYNQ module, and converts the signals other than vibration into digital signals and communicates them to the DSP module.

[0079] The data acquisition module can achieve high-speed, high-precision, and high-real-time acquisition of the analog signals of the engine room equipment to be monitored and convert them into digital signals. Then, the core board module caches and processes the data, and transmits the signal data to the upper computer through the data communication module, avoiding data distortion and improving the accuracy and efficiency of data acquisition and processing.

[0080] The data acquisition module processes the analog signals in the engine room and converts them into digital signals, and then transmits them to the core board module, including: the ZYNQ module, which communicates with the A / D analog-to-digital conversion module to receive vibration signals, stores the high-frequency vibration signals in the DDR3 of the data buffer module through the AXI bus and FIFO (first in first out), and transmits the vibration signals to the intelligent management unit through the RJ gigabit Ethernet port and the UDP protocol of the data communication module after reading the buffered data; the DSP module, which receives the vibration signals transmitted by the ZYNQ module and the non-vibration signals of the equipment to be monitored in the engine room, writes the signals into the signal processing module, and then transmits the processed signals to the intelligent management unit through the RS communication module.

[0081] Specifically, the core board module further includes an SPI communication module composed of a ZYNQ module and a DSP module, which transmits the vibration signals collected by the ZYNQ module to the DSP module through a multi-channel SPI communication method of one master and multiple slaves; SPI communication is a full-duplex synchronous serial communication protocol that allows data to be sent and received simultaneously, enabling high-speed data transmission. There are a total of four signal lines, namely: MOSI (master output slave input), MISO (master input slave output), SCK (master device generates clock signal), and CS (chip select signal). The working mode is determined by the clock polarity (CPOL) and clock phase (CPHA). By combining the two, the sampling timing is selected. In the core board module, it is a single-master multi-slave mode, where the ZYNQ module is the master, and each channel of the DSP module that receives vibration signals is a slave. After the ZYNQ module master pulls down the chip select signal, the DSP module receives the master device clock signal SCK and the vibration signal slave input data MOSI. The working mode is that both the clock polarity (CPOL) and clock phase (CPHA) are low levels, that is, the MOSI electrical signal is read when the clock signal SCK is pulled high, and then the received vibration signal is sent to the signal processing module for processing.

[0082] Specifically, the ZYNQ module includes: an AXI bus transmission module, a data cache module DDR3, and a UDP protocol module. When electrical signals are written into the data cache module DDR3 in the ZYNQ module and when data is transmitted to the intelligent management unit to read the data cache module DDR3, both rely on the AXI bus protocol. AXI is a high-performance, low-latency, and flexible and scalable on-chip bus standard, suitable for various high-bandwidth and high-concurrency data transmission requirements; for the data cache module DDR3, after the ZYNQ module receives the signal sent by the A / D analog-to-digital conversion module, due to the large amount and high rate of vibration signal data, the data is put into the data cache module DDR3 through the AXI bus transmission module to prevent data loss. When the ZYNQ module transmits data to the intelligent management unit through the UDP protocol module, the vibration signal cached in the data cache module DDR3 is read through the AXI bus transmission module again. The UDP protocol module is suitable for transmitting large amounts of vibration signal data when the ZYNQ module communicates with the intelligent management unit, and sends the collected data to the management unit through the gigabit Ethernet port.

[0083] Specifically, the DSP module includes: a signal processing module, which is suitable for the DSP module to receive the vibration signal sent by the ZYNQ module and collect the non-vibration signal of the equipment to be monitored in the engine room, and then process the signal. Specifically, after calculating the one-dimensional time-series characteristics of the vibration signal and the actual physical state information of the non-vibration signal, it is sent to the intelligent management unit for fault diagnosis.

[0084] Specifically, the hardware acquisition unit also includes a data communication module, which is controlled by the core board module and displays the processed engine room signals in the intelligent management unit. Among them, the high-rate vibration signal is transmitted through the gigabit Ethernet port and the UDP protocol, and other non-vibration signals are transmitted through the serial-to-USB and RS protocols.

[0085] Furthermore, the intelligent management unit includes: an online monitoring module, a data storage module, a fault diagnosis module, and an intelligent operation and maintenance module. The online monitoring module is connected and communicates with the core board module, receives the signals transmitted by the ZYNQ module and the DSP module, and after relevant processing, obtains the corresponding engine room equipment signals, so as to realize the monitoring of the operating status of the equipment to be monitored in the engine room; the data storage module is communicatively connected to the database and the core board module, and is suitable for storing its signal data into the database when sudden situations such as faults occur in the equipment to be monitored in the engine room, and performing periodic storage of daily data. The fault diagnosis module is communicatively connected to the database and the core board module, analyzes the signals fed back by the equipment to be monitored in the engine room through an intelligent algorithm model, and diagnoses whether a fault has occurred and the degree of the fault; the intelligent operation and maintenance module, after the fault diagnosis module judges whether a fault has occurred and the health status of the equipment according to the operating data of the equipment to be monitored in the engine room, the intelligent operation and maintenance module makes an evaluation based on the results and gives operation and maintenance suggestions to the staff.

[0086] The above-mentioned fault diagnosis module performs time-frequency domain analysis and processing on the collected data signals through intelligent processing algorithms. Based on the professional knowledge of fault diagnosis of electrical equipment, propulsion equipment, and cooling equipment, it determines whether there is a fault in the engine room equipment and gives an alarm prompt. The health assessment module evaluates the degree of fault of the engine room equipment based on the results of the fault diagnosis and puts forward reasonable maintenance suggestions according to different health levels.

[0087] Among them, the intelligent fault diagnosis algorithm is a fault diagnosis algorithm based on the Markov Transition Field (MTF) and the Reparameterized Local Kernel Networks (RepLKNet). Its steps are as Figure 3 shown.

[0088] First, the Markov Transition Field (MTF) converts the one-dimensional time series signal into two-dimensional time-frequency image features. It is assumed that the transition matrix M of the system with N discrete states is an N×N matrix, where each element represents the transition probability from state to state . If this matrix is defined by the Q quantile, then its dimension is Q×Q.

[0089] Specifically, the Markov Transition Field (MTF) assumes a time series as follows:

[0090]

[0091] where each is the state at time, and is the length of the sequence.

[0092] Then, the time series is divided into N quantile segments, labeled 1, 2,..., N. Then, each data in is corrected to the corresponding quantile segment label. Subsequently, the Markov state transition matrix W is constructed. The element in the th row and th column represents the probability that the quantile segment transfers to :

[0093]

[0094] Based on this state transition matrix, the MTF is given as:

[0095]

[0096] The numerical value of MTF is directly imaged into a grayscale image or mapped into a pseudo-color image, that is, the transformation from a one-dimensional time series signal to an image is realized, and two-dimensional time-frequency features are obtained.

[0097] Next, the one-dimensional time series features are combined with the two-dimensional time-frequency features obtained by the Markov transition field MTF transformation, and the features are respectively input into the neural network RepLKNet based on large convolutional kernels for training, and the computational efficiency and classification performance are improved through local convolutional kernels and reparameterization techniques.

[0098] The features are fused by splicing. Feature splicing means that features from different modalities (one-dimensional time series features and two-dimensional image features) are fused and connected together through the channel dimension, so as to form a comprehensive feature representation. In the RepLKNet neural network, splicing is usually performed by channel or by dimension.

[0099] Among them is the one-dimensional time series feature, with a shape of ( , ), where is the number of samples, is the feature length of each sample; is the two-dimensional image feature, with a shape of ( , , , ), where is the number of samples, , are the height and width of the image respectively, is the number of channels of the image.

[0100] These two types of features (time series features and image features) are spliced to generate a composite feature representation for classification.

[0101] One-dimensional time series feature can extract local features through some convolutional operations or directly convert them into a vector with a lower dimension through a fully connected layer. Assuming it is converted into a feature vector with dimensions:

[0102]

[0103] Among them , is the convolutional operation, is the convolutional kernel.

[0104] Two-dimensional image feature After being processed by the local convolutional operation in RepLKNet in the convolutional neural network, it obtains a shape of ( , , , The tensor of ). To splice with one-dimensional features, it is usually necessary to perform pooling or convolution on the image features to adjust their dimensions. Assume the processed image features are , and its shape is .

[0105]

[0106] Among them , is the pooling operation.

[0107] Splice the processed one-dimensional time series features and two-dimensional image features on the feature dimension to obtain a new feature representation :

[0108]

[0109] The spliced features can be passed into a local convolutional layer to extract finer-grained local features. You can pass the spliced features as input to the convolutional layer to further learn local patterns.

[0110] The neural network RepLKNet based on large convolutional kernels introduces large convolutional kernels, which can more effectively capture long-range context information. Among them, the large convolutional kernels are reparameterized, usually by decomposing larger convolutional kernels into smaller convolutional operations (such as 1x1 convolutions and other local convolutions) to reduce computational overhead. The reparameterized convolution can be represented in a form similar to the following:

[0111]

[0112] Among them, is the convolutional kernel after reparameterization. The core idea is to decompose complex convolutional operations into a set of simple and more efficient operations.

[0113] The neural network RepLKNet based on large convolutional kernels uses local convolutional kernels to replace traditional large convolutional kernels. Assume there is a local convolutional kernel , and the output of this local convolution can be expressed as:

[0114]

[0115] Among them is the local convolutional kernel. Local convolution captures local features by restricting the size of the convolutional kernel to a smaller local area, thus reducing the amount of computation.

[0116] The neural network RepLKNet based on large convolutional kernels extracts features at different levels through convolutional operations of multiple scales. Generally, the network starts with smaller local convolutions and then aggregates these features through global convolution or pooling operations. Let the convolutional operations at different scales be:

[0117]

[0118] Finally, the trained concatenated features are input into the fully connected layer for classification operations:

[0119]

[0120] Among them, is the final classification output, that is, the output fault category, is the weight matrix of the fully connected layer, is the bias term, The function is used to map the output to the class probability space.

[0121] During the training process, the neural network RepLKNet based on large convolutional kernels adopts the technology of adaptive noise to enhance the robustness of local convolutions. Noise is added to the input and its intensity is adjusted in each training iteration. The input after adding noise is expressed as:

[0122]

[0123] Among them, is the noise generated according to needs, usually gradually decreasing during the training process. In this way, the neural network RepLKNet based on large convolutional kernels can increase robustness and reduce mode aliasing when processing signals.

[0124] The loss function of the neural network RepLKNet based on large convolutional kernels is usually based on common classification loss functions, such as cross-entropy loss. In the classification task, the cross-entropy loss function is:

[0125]

[0126] Among them, is the true label, is the predicted probability output by the network, is the number of classes, e is the natural logarithm, and using the natural logarithm helps to calculate the gradient during the optimization process and avoid unnecessary complexity.

[0127] Example 2

[0128] This example discloses a method for processing ship power data. Using the above-mentioned ship power data processing system, the method for processing ship power data includes:

[0129] Step S101: Collect the analog signals of the equipment to be monitored in the engine room through the hardware acquisition unit;

[0130] Step S103: Perform analog-to-digital conversion and processing on the analog signals of the equipment, and send the processing result to the upper computer intelligent management unit after obtaining the result;

[0131] Step S105: The intelligent management unit realizes on-line monitoring of the engine room equipment, and uses intelligent algorithms to diagnose faults on the parsed vibration signal data, and obtains a conclusion on whether a fault has occurred;

[0132] Step S107: Give intelligent operation and maintenance suggestions to relevant personnel according to the conclusion;

[0133] Step S109: When sudden situations such as faults occur in the equipment to be monitored in the engine room, store its signal data in the database, and perform periodic storage on the daily data.

[0134] Furthermore, in step S103, the analog signals of the operating state are processed, and after obtaining the data processing result, it is sent to the upper computer intelligent management unit;

[0135] Step S201: Collect the status information of the engine room equipment through the sensor and output analog signals;

[0136] Step S203: Collect the analog signals and perform operations such as amplification, filtering, conversion, and isolation on the analog signals;

[0137] Step S205: Convert the processed analog signals into digital signals through analog-to-digital conversion and send them to the core board module;

[0138] Step S207: The ZYNQ module in the core board caches and transmits the received high-speed vibration signals to the upper computer intelligent management unit, and communicates the vibration signals to the DSP module. At the same time, the DSP module receives the external vibration signals, processes the signals, and then sends them to the upper computer intelligent management unit.

[0139] By designing a high-speed signal acquisition and processing module based on the ZYNQ module and the DSP module, different types of signals in the engine room are collected separately and preliminarily processed, which improves the integrity and real-time performance of signal acquisition and effectively avoids the distortion of high-frequency signals.

[0140] Even further, in step S105, on-line monitoring of the engine room equipment is realized, and the parsed vibration signal data is used for fault diagnosis with intelligent algorithms, specifically including:

[0141] Step S301: Receive the transmission signals from the ZYNQ module and the DSP module, perform relevant processing, and obtain the corresponding physical state information, thereby realizing the online monitoring of the operating state of the equipment to be monitored in the engine room;

[0142] Step S302: Analyze the signals fed back by the equipment to be monitored in the engine room through an intelligent algorithm model, and diagnose whether a fault has occurred and the degree of the fault.

[0143] The intelligent diagnosis algorithm is implemented through a training model. After signal processing of the collected vibration signals, signal features are extracted, and the processed content is input into the trained intelligent diagnosis algorithm model to obtain a more reliable and accurate fault assessment conclusion.

[0144] The ship power data processing method further includes:

[0145] Step S401: When a fault occurs, store the fault data and nodes in the database, and periodically store the daily data;

[0146] Step S403: Query and access the data records and engine room logs in the database.

[0147] Furthermore, the ship power data processing method further includes:

[0148] Step S501: Perform a cycle every two days to back up the database and update the data stored in the database.

[0149] Among them, the cycle time for database backup and update can also be selected with other intervals, which should be reasonably set according to actual needs to facilitate subsequent research and processing of the faulty parts.

[0150] Embodiment 3

[0151] A computer storage medium, a hardware device suitable for storing data and programs, which implements the steps of the ship power data processing method in Embodiment 2 when the program and data are called.

[0152] Those skilled in the art can understand that to implement all or part of the processes in the above embodiment methods, it can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above method embodiments. The storage medium can be selected as:

[0153] 1) Main memory (memory)

[0154] RAM (Random Access Memory): Used to temporarily store data and programs, with relatively fast read and write speeds, but the data will be lost after power failure;

[0155] ROM (Read Only Memory): Used to store firmware programs or data. The data is not easily changed and will not be lost during power outages.

[0156] 2) Auxiliary storage

[0157] Hard disk (HDD): Mechanical storage medium with large capacity but slow reading and writing speed;

[0158] Solid-state drive (SSD): Based on flash memory technology, it has fast read and write speed, large capacity, and strong shock resistance;

[0159] Optical disc (CD / DVD): used for data storage and transmission, with slow reading and writing speed and limited capacity;

[0160] USB flash drive: a portable storage device with faster read and write speeds than a hard drive and moderate capacity;

[0161] 3) Network storage

[0162] NAS (Network Attached Storage): Shares storage resources over a network and is suitable for multiple devices to access together;

[0163] Cloud storage: Storing data on remote servers that users can access via the internet.

[0164] Flexibly select storage media based on storage capacity, access speed, and data.

[0165] Example 4

[0166] An electronic device comprises at least a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program in the memory, the steps of the ship power data processing method of embodiment 2 are performed.

[0167] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the invention.

Claims

1. A ship power data processing system, characterized in that, Comprising: A hardware acquisition unit, which is suitable for acquiring and processing the analog quantity data of the equipment to be monitored in the engine room, and transmitting the real-time operation state signal of the engine room equipment collected to the intelligent management unit; The intelligent management unit converts the collected real-time operation state signal into corresponding physical state information, and at the same time uses an intelligent diagnosis algorithm to diagnose faults in the vibration signal, and judges whether there is a fault in the engine room based on the diagnosis result; A database, which is suitable for storing the signal data into the database when it is detected that there is a fault in the equipment to be monitored in the engine room, and storing the normal data periodically; The intelligent diagnosis algorithm is a fault diagnosis algorithm based on a Markov transition field and a neural network based on a large convolutional kernel, and its operation steps are as follows: First, use the Markov transition field MTF to convert the collected one-dimensional time series signal into two-dimensional time-frequency image features; Then, input the one-dimensional time series signal and two-dimensional time-frequency image features into the neural network RepLKNet based on large convolution kernels. In the neural network RepLKNet based on large convolution kernels, perform concatenated feature fusion on the one-dimensional time series features and two-dimensional image features by dimension, and form a comprehensive feature representation. Then, input the comprehensive feature representation into a local convolution layer to extract finer-grained local features; the neural network RepLKNet extracts features at different levels through convolution operations at multiple scales and trains. Finally, input the trained concatenated features into a fully connected layer for classification operations, that is, output the fault category; in the neural network RepLKNet based on large convolution kernels, perform concatenated feature fusion on the one-dimensional time series features and two-dimensional image features by dimension, and form a comprehensive feature representation including: set X 1D is the one-dimensional time series feature, with a shape of (N 1D , T 1D ), where N 1D is the number of samples, and T 1D is the feature length of each sample; X 2D is the two-dimensional image feature, with a shape of (N 2D , H 2D , W 2D , C 2D ), where N 2D is the number of samples, H 2D , W 2D are the height and width of the image respectively, and C 2D is the number of channels of the image; The one-dimensional time series feature X 1D Extract local features through a convolution operation or directly convert it into a low-dimensional vector through a fully connected layer, and convert it into a feature vector with F 1D dimensions: X′ 1D = Conv(X 1D , K); Among them Conv is the convolution operation, and K is the convolution kernel; Two-dimensional image feature X 2D After being processed by the local convolution operation in the neural network RepLKNet with a large convolution kernel in the convolutional neural network, a tensor with the shape (N, H 2D , W 2D , C 2D ) is obtained; and the image feature is pooled or convolved to adjust its dimension, and the processed image feature is set as X' 2D , and its shape is N×F 2D ; X′ 2D = pool(X 2D ); wherein pool is a pooling operation; The processed one-dimensional time series feature X′ 1D and the two-dimensional image feature X′ 2D are concatenated in the feature dimension to obtain a new feature representation X fused : X fused = concat(X' 1D , X' 2D ).

2. The ship power data processing system according to claim 1, characterized in that, The conversion of the collected one-dimensional time series signal into two-dimensional time-frequency image features by using the Markov transition field MTF includes: The transition matrix M that sets the system to have N discrete states is an N×N matrix, where each element M ij represents the transition probability from state i to state j; the transition matrix is defined using the Q quantile; Specifically, set the Markov transition field MTF time series as follows: X = {x1, x2,..., x T}; Where each x is the state at each moment, and T is the length of the sequence; Then divide the time series X into N quantile segments, labeled 1, 2,..., N, and then correct each data in X to the corresponding quantile segment label; subsequently construct the Markov state transition matrix W, and represent the element in the i-th row and j-th column as the probability of the quantile segment i transitioning to j: Based on this state transition matrix, give the Markov transition field MTF: Directly image the values of the Markov transition field MTF into a grayscale image or map it to a pseudo-color image, that is, realize the transformation of the one-dimensional time series signal into an image, and obtain two-dimensional time-frequency features.

3. The ship power data processing system according to claim 1, wherein The neural network RepLKNet based on a large convolutional kernel extracts features at different levels through convolutional operations at multiple scales, and finally inputs the trained concatenated features into a fully connected layer for classification operations, and the output fault categories include: The neural network RepLKNet based on a large convolutional kernel extracts features at different levels through convolutional operations at multiple scales; the neural network starts with small local convolutions, and then aggregates features at different levels through global convolution or pooling operations; among them, the convolutional operations at different scales are: y1 = K1 * x, y2 = K2 * x,..., y n = K n * x; Finally, input the concatenated features into a fully connected layer for classification operations: Among them, is the final classification output, that is, the output fault category, W fc is the weight matrix of the fully connected layer, b fc is the bias term, and the softmax function is used to map the output to the category probability space; During the concatenated training process, the neural network RepLKNet based on a large convolutional kernel adopts an adaptive noise algorithm to enhance the robustness of local convolutions; noise is added to the input, and its intensity is adjusted in each training iteration; the input x' after adding noise is expressed as: x' = x + η; Where η is the noise generated as needed, and it gradually decreases during the training process; the loss function of the neural network RepLKNet based on a large convolutional kernel is the cross-entropy loss function. In the classification task, the cross-entropy loss function is: where y i is the true label, is the predicted probability output by the network, c is the number of classes, and e is the natural logarithm.

4. The ship power data processing system according to claim 1, characterized in that, The equipment to be monitored in the engine room includes engines, pumps, compressors, bearings and couplings; the hardware acquisition unit includes: a data acquisition module, which is suitable for acquiring the vibration signals of the equipment to be monitored in the engine room; an A / D analog-to-digital conversion module, which monitors the analog signals of the equipment to be monitored in the engine room through sensors, and after being processed by the signal conditioning module, converts the acquired vibration signals into digital signals and communicates them to the ZYNQ module, and converts the signals other than vibration into digital signals and communicates them to the DSP module.

5. A method for processing ship power data, characterized in that, Adopt the ship power data processing system according to any one of claims 1 to 4, and the ship power data processing method includes: Use the designed core board module as a high-speed acquisition card and a control core to acquire the signals of the equipment to be monitored in the engine room; Perform corresponding processing on the acquired analog signals in the engine room and transmit them to the upper computer; Use intelligent algorithms to diagnose faults in the processed vibration signal data to determine whether a fault has occurred; Give intelligent operation and maintenance suggestions to relevant personnel according to the conclusion; When a sudden failure occurs in the equipment to be monitored in the engine room, store its signal data in the database, and periodically store the daily data.

6. A computer storage medium, characterized in that, It includes a hardware device for storing data and programs, and when the programs and data are called, the steps of the ship power data processing method described in claim 5 are implemented.

Images