A marine internal combustion engine fault monitoring instrument and system based on AI technology

Abstract

This invention discloses a marine internal combustion engine fault monitoring instrument and system based on AI technology, belonging to the field of fault monitoring technology. It includes a multi-source sensor cluster, an edge AI processor, and a shipborne monitor. The edge AI processor is fixedly installed on the top of the outer wall of the multi-source sensor cluster, and the shipborne monitor is fixedly installed on the surface of the internal combustion engine cylinder block. The edge AI processor is connected to the multi-source sensor cluster via a data cable to establish a bidirectional data channel. The shipborne monitor internally houses a vibration sensing substrate and a temperature conducting copper tube. A piezoelectric ceramic array is disposed on the top of the outer wall of the vibration sensing substrate, and the piezoelectric ceramic array is embedded in the inner wall of the shipborne monitor in an equilateral triangle arrangement. This invention, through the multi-source sensor cluster, achieves simultaneous multi-dimensional data acquisition of marine internal combustion engines, solving the problem of single monitoring methods and inability to comprehensively capture fault characteristics, thus enabling earlier and more accurate prediction of internal combustion engine faults.

Description

Technical Field

[0001] This invention relates to the field of fault monitoring technology, specifically to a fault monitoring instrument and system for marine internal combustion engines based on AI technology. Background Technology

[0002] As the core power unit of a ship, the operating status of the internal combustion engine directly affects the safety and transportation efficiency of the ship. Existing fault monitoring methods have problems such as poor real-time performance and insufficient diagnostic accuracy, which makes it difficult to meet the stringent requirements of modern shipping for the reliability of the power system. With the deep application of artificial intelligence technology in the industrial field, we are trying to integrate it into the monitoring equipment of the internal combustion engine of the ship.

[0003] Patent CN110708114B discloses a method and system for locating and visualizing optical cable line faults based on AI image recognition. The above patent achieves accurate location of optical cable fault points and can clearly identify the type and cause of the optical cable fault at the same time, which facilitates maintenance personnel to quickly arrive at the site, allows maintenance personnel to prepare maintenance tools and solutions in advance, improves maintenance efficiency, and reduces losses.

[0004] The aforementioned patents enable precise location of optical cable line faults, clearly presenting the fault type and cause, greatly facilitating the work of maintenance personnel. This allows them to prepare maintenance tools and solutions in advance, effectively improving maintenance efficiency and reducing losses caused by faults. However, the patents do not focus on monitoring faults in marine internal combustion engines.

[0005] Therefore, this application proposes a marine internal combustion engine fault monitoring instrument and system based on AI technology, which can collect multi-source data, perform noise reduction through feature extraction components, and analyze and diagnose the fault feature library and LSTM operation layer to obtain corresponding maintenance methods. Summary of the Invention

[0006] The purpose of this invention is to provide a marine internal combustion engine fault monitoring instrument and system based on AI technology, so as to solve the technical problems of insufficient monitoring capability for changing operating conditions and difficulty in data processing of existing monitoring methods mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a marine internal combustion engine fault monitoring instrument based on AI technology, comprising a multi-source sensor cluster, an edge AI processor, and a shipborne monitor. The edge AI processor is fixedly installed on the top of the outer wall of the multi-source sensor cluster, and the shipborne monitor is fixedly installed on the surface of the internal combustion engine cylinder block. The edge AI processor is connected to the multi-source sensor cluster via a data cable to establish a bidirectional data channel.

[0008] The shipborne monitor is equipped with a vibration sensing substrate and a temperature conducting copper tube. A piezoelectric ceramic array is set at the top of the outer wall of the vibration sensing substrate. The piezoelectric ceramic array is embedded in the inner wall of the shipborne monitor in an equilateral triangle arrangement. The temperature conducting copper tube extends through the shipborne monitor to the threaded mounting hole of the internal combustion engine cylinder block.

[0009] Preferably, the multi-source sensor cluster is fixedly installed on the inner wall of the housing by a bracket, and the multi-source sensor cluster is inserted into the heterogeneous interface on the outer wall of the edge AI processor by a cable bundle. The multi-source sensor cluster includes a vibration sensor array, an oil monitoring probe, a temperature sensor group and a crankshaft phase sensor. The vibration sensor array is fixedly installed on the inner wall of the cylinder by a bracket and connected to the cylinder head bolts of the internal combustion engine. The flow channel interface of the oil monitoring probe is vertically inserted into the main return oil pipe of the lubricating oil, and the quartz sensing cavity of the oil monitoring probe is welded to the oil pipe through a flange.

[0010] The temperature sensor group includes a thermocouple and an infrared thermometer. The probe end of the thermocouple is nested in the internal thread of the temperature measuring hole in the exhaust manifold. The infrared thermometer is attached to the elbow of the exhaust pipe by a magnetic base. The Hall chip of the crankshaft phase sensor is fitted with the end face of the flywheel gear ring. The Hall chip carrier is fixed to the edge of the crankshaft by bracket bolts.

[0011] Preferably, the edge AI processor includes an acceleration board, an AI inference chip, and a feature extraction component. The acceleration board is welded to the top of the outer wall of the heat-conducting substrate via a ball grid array. The heat-conducting substrate is fixedly installed on the inner wall of the ship's control cabin. The AI ​​inference chip is welded to the top of the outer wall of the acceleration board, and the feature extraction component is vertically inserted into the top of the outer wall of the AI ​​inference chip.

[0012] The top of the outer wall of the accelerator board is welded with a dynamic reconfiguration pin array, and four sets of memory chips surround the edge area. The memory chips are connected in series by micro data lines.

[0013] Preferably, the feature extraction component comprises two physically isolated parts, namely a vibration feature processing unit in the left half and an oil feature processing unit in the right half. An electromagnetic shielding plate is provided between the vibration feature processing unit and the oil feature processing unit. The vibration feature processing unit is connected to an anti-aliasing filter via gold wire bonding, and the oil feature processing unit is connected to an AI inference chip via a data cable.

[0014] Preferably, the AI ​​inference chip is configured with a lower LSTM computing layer and an upper fault feature library layer. The fault feature library layer is bonded to the LSTM computing layer through microbumps. The LSTM computing layer is electrically connected to the PCIe slot of the acceleration board through through-silicon vias to form a conductive path. The flash memory array of the fault feature library layer is connected to the input pin of the vibration feature processing unit through a data line.

[0015] A data cable runs through the outer side of the AI ​​inference chip, connecting to a temperature sensing ring.

[0016] Preferably, the oil monitoring probe includes an oil quality sensor and an ultrasonic viscosity sensor. The oil quality sensor is wound around the outer wall of the base tube, and the piezoelectric transducer ring of the ultrasonic viscosity sensor is coaxially sleeved on the outer wall of the base tube. The data line outlet end of the piezoelectric transducer ring forms a contact. The lead wire of the oil quality sensor passes through the inner hole of the base tube and its end is connected to a micro socket. The micro socket and the contact are inserted into the dual-channel slot of the connector.

[0017] An aviation interface group is set on the outer side of the shell, and a satellite communicator is fixedly installed on the top of the inner wall of the shell through a high-frequency isolation plate. A power manager is set in the interlayer of the inner side of the shell.

[0018] Preferably, the output of the vibration sensor array is connected to the ADC interface of the feature extraction component via a data cable, the oil monitoring probe and temperature sensor group are connected to the AI ​​inference chip via a multiplexer, the AI ​​inference chip is interconnected with the storage library via a slot, and the network interface of the AI ​​inference chip is connected to the satellite communicator via a data cable.

[0019] Preferably, the power manager has a surge protection circuit and a battery redundancy switching module stacked inside. The input terminal of the surge protection circuit passes through the housing and is connected to the ship's main power line, and the output terminal of the surge protection circuit is vertically connected to the battery redundancy switching module through a copper busbar.

[0020] The battery redundancy switching module includes a main battery compartment and a backup battery compartment. An electromagnetic relay is installed between the main battery compartment and the backup battery compartment. The moving contact of the armature of the electromagnetic relay is connected to the output bus through a phosphor bronze spring.

[0021] Preferably, the system includes an AI model training server and a fault decision engine. The AI ​​inference chip and the AI ​​model training server form a closed loop. The AI ​​model training server is connected to the knowledge base output end through a fiber optic switch. The diagnostic results of the edge AI processor are connected to the input end of the fault decision engine. The output end of the fault decision engine is connected to the maintenance plan generator. The satellite communicator establishes a two-way link with the ship-shore communication network. The ship-shore communication network transmits data in parallel through a maritime satellite channel and a redundant 4G / 5G channel.

[0022] Preferably, the AI ​​model training server has a built-in LSTM optimization algorithm. The parameters output by the AI ​​model training server are encrypted and compressed and then transmitted to the fault feature library. The AI ​​inference chip is marked with a timestamp watermark for the fault sample data and then sent back to the knowledge base. The vibration sensor array collects the cylinder head vibration signal, the oil monitoring probe obtains the dielectric constant of the lubricating oil, the temperature sensor group monitors the exhaust temperature, the vibration feature processing unit performs wavelet noise reduction, the oil feature processing unit performs pulse coding, and the accelerator board accelerates the preprocessed data in parallel before inputting it into the AI ​​inference chip through the PCIe slot.

[0023] The LSTM computation layer calls the model parameters from the fault feature library and outputs the fault probability matrix. The satellite communicator splits the diagnostic data packet into: core feature data transmitted in real time via the maritime satellite channel and raw waveform data transmitted asynchronously via the 4G / 5G channel. The fault decision engine aligns the core feature data and raw waveform data in time and space, and generates three levels of maintenance instructions by combining them with the maintenance knowledge graph: Level 1 instructions are directly issued to the display terminal of the shipborne monitor; Level 2 instructions trigger the pre-delivery of spare parts; and Level 3 instructions are sent to the shore-based expert consultation center.

[0024] Compared with the prior art, the beneficial effects of the present invention are:

[0025] 1. This invention achieves the function of synchronous acquisition of multi-dimensional data of marine internal combustion engines through a multi-source sensor cluster, which solves the problem of single monitoring methods and inability to fully capture fault characteristics, and predicts internal combustion engine faults earlier and more accurately;

[0026] 2. This invention enables real-time AI analysis and fault diagnosis of multi-source sensor data on the ship via an edge AI processor, solving the problem of inability to diagnose in a timely manner when relying on shore-based data centers due to transmission delays and communication interruptions, thereby improving the real-time performance and reliability of fault diagnosis.

[0027] 3. This invention uses an AI inference chip to achieve time-series data processing and fault mode recognition based on deep learning, which solves the problem that traditional algorithms have difficulty in handling the complex dynamic characteristics and nonlinear fault features of internal combustion engines, thereby improving the accuracy of fault diagnosis and reducing the false alarm rate.

[0028] 4. This invention establishes a two-way link between a satellite communicator and a ship-to-shore communication network, realizing dual-channel redundant communication functions for real-time transmission of core feature data via maritime satellite channels and asynchronous transmission of raw waveform data via 4G / 5G channels. This solves the problem of insufficient reliability caused by a single method under complex sea conditions, ensuring reliable transmission of fault data and timely remote guidance from shore-based experts. Attached Figure Description

[0029] Figure 1 This is a front view structural diagram of the present invention;

[0030] Figure 2 This is a schematic diagram of the internal structure of the edge AI processor of the present invention;

[0031] Figure 3 This is a schematic diagram of the internal structure of the housing of the present invention;

[0032] Figure 4 This is a schematic diagram of the shipborne monitoring device of the present invention;

[0033] Figure 5 This is a schematic diagram of the power manager structure of the present invention;

[0034] Figure 6 This is a schematic diagram of the internal structure of the multi-source sensor cluster of the present invention;

[0035] Figure 7 This is a schematic diagram of the internal structure of the oil monitoring probe of the present invention;

[0036] Figure 8 This is a schematic diagram of the internal structure of the AI ​​model training server of the present invention;

[0037] Figure 9 This is a schematic diagram of the internal structure of the temperature sensor group of the present invention.

[0038] In the diagram: 1. Multi-source sensor cluster; 2. Edge AI processor; 3. Shipborne monitor; 4. Vibration sensing substrate; 5. Temperature conduction copper tube; 6. Piezoelectric ceramic array; 7. Housing; 8. Vibration sensor array; 9. Oil monitoring probe; 10. Temperature sensor group; 11. Crankshaft phase sensor; 12. Thermocouple; 13. Infrared temperature measuring head; 14. Accelerator plate; 15. AI inference chip; 16. Feature extraction component; 17. Vibration feature processing unit; 18. Oil feature processing unit; 19. Temperature sensing ring; 20. Oil quality sensor; 21. Ultrasonic viscosity sensor; 22. Aviation interface group; 23. Satellite communicator; 24. Power manager; 25. Base tube; 26. Battery redundancy switching module; 27. Main battery compartment; 28. Backup battery compartment; 29. ​​Electromagnetic relay; 30. AI model training server; 31. Fault decision engine; 32. Maintenance plan generator. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," "outer," "front end," "rear end," "both ends," "one end," and "the other end," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0041] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installed," "equipped with," "connected," etc., should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0042] Please see Figure 1 , Figure 3 , Figure 4 , Figure 6 , Figure 7 and Figure 9 The present invention provides an embodiment of a marine internal combustion engine fault monitoring instrument based on AI technology, comprising a multi-source sensor cluster 1, an edge AI processor 2, and a shipborne monitor 3. The edge AI processor 2 is fixedly installed on the top of the outer wall of the multi-source sensor cluster 1, and the shipborne monitor 3 is fixedly installed on the surface of the internal combustion engine cylinder block. The edge AI processor 2 is connected to the multi-source sensor cluster 1 through a data cable to establish a bidirectional data channel.

[0043] The multi-source sensor cluster 1 is fixedly installed on the inner wall of the housing 7 by a bracket. The multi-source sensor cluster 1 is inserted into the heterogeneous interface on the outer side of the edge AI processor 2 through a cable bundle. The multi-source sensor cluster 1 includes a vibration sensor array 8, an oil monitoring probe 9, a temperature sensor group 10, and a crankshaft phase sensor 11. The vibration sensor array 8 is fixedly installed on the inner wall of the cylinder by a bracket and connected to the cylinder head bolt of the internal combustion engine. The flow channel interface of the oil monitoring probe 9 is vertically inserted into the main return oil pipe of the lubricating oil. The quartz sensing cavity of the oil monitoring probe 9 is welded to the oil pipe through a flange. The temperature sensor group 10 includes a thermocouple 12 and an infrared temperature measuring head 13. The detection end of the thermocouple 12 is nested in the internal thread of the temperature measuring hole of the exhaust manifold. The infrared temperature measuring head 13 is attracted to the elbow of the exhaust pipe by a magnetic base. The Hall chip of the crankshaft phase sensor 11 is fitted with the end face of the flywheel gear ring. The Hall chip carrier is fixed to the edge of the crankshaft by a bracket bolt.

[0044] Furthermore, the sensitive axis of the piezoelectric ceramic probe is installed at a 45° angle to the cylinder head plane, the oil quality sensor 20 and the ultrasonic viscosity sensor 21 are coaxially wound on the outer wall of the base tube 25, the thermocouple 12 of the temperature sensor group 10 is nested in the threaded hole of the exhaust manifold, and the infrared temperature measuring head 13 is magnetically attracted to the elbow of the exhaust pipe.

[0045] When the internal combustion engine is running, the piezoelectric ceramic probe of the vibration sensor array 8 captures the three-dimensional vibration signal of the cylinder head. The 45° angle can simultaneously sense the radial and axial vibration components. The vibration signal is transmitted to the heterogeneous interface of the edge AI processor 2 through the shielded cable and processed by the vibration feature processing unit 17 of the feature extraction component 16: first, high-frequency noise >20kHz is removed by the anti-aliasing filter with gold wire bonding.

[0046] Then, the microcontroller executes a wavelet noise reduction algorithm to extract the characteristic frequency band of 0.5-5kHz, which corresponds to the piston ring wear characteristic frequency; a 10MHz AC electric field is applied to the spiral electrode of the oil monitoring probe 9, and the concentration of lubricating oil metal particles is detected by the change in dielectric constant; the ultrasonic viscosity sensor 21 emits a 5MHz pulse wave and calculates the viscosity value based on the echo delay difference; the two signals are input to the ADC chip through a flexible circuit and converted into digital quantities, which are then compressed to 30% of the original data by a Huffman encoder; the temperature sensor group 10 synchronously collects the exhaust manifold temperature gradient, and the data from the thermocouple 12 and the infrared temperature measuring head 13 are input to the AI ​​inference chip 15 through a multiplexer.

[0047] Please see Figure 2 , Figure 3 and Figure 4 The present invention provides an embodiment of a marine internal combustion engine fault monitoring instrument based on AI technology. The edge AI processor 2 includes an acceleration plate 14, an AI inference chip 15, and a feature extraction component 16. The acceleration plate 14 is welded to the top of the outer wall of a heat-conducting substrate via a ball grid array. The heat-conducting substrate is fixedly installed on the inner wall of the ship's control cabin. The AI ​​inference chip 15 is welded to the top of the outer wall of the acceleration plate 14, and the feature extraction component 16 is vertically inserted into the top of the outer wall of the AI ​​inference chip 15. A dynamic reconstruction pin array is welded to the top of the outer wall of the acceleration plate 14, and four sets of storage particles surround the edge region. The storage particles are connected in series via micro data lines.

[0048] The feature extraction component 16 comprises two physically isolated parts: a vibration feature processing unit 17 in the left half and an oil feature processing unit 18 in the right half. An electromagnetic shielding plate is provided between the vibration feature processing unit 17 and the oil feature processing unit 18. The vibration feature processing unit 17 is connected to an anti-aliasing filter via gold wire bonding, and the oil feature processing unit 18 is connected to the AI ​​inference chip 15 via a data line. The AI ​​inference chip 15 is configured as a lower LSTM operation layer and an upper fault feature library layer. The fault feature library layer is bonded to the LSTM operation layer via microbumps. The LSTM operation layer is electrically connected to the PCIe slot of the accelerator board 14 via through-silicon vias to form a conductive path. The flash memory array of the fault feature library layer is connected to the input pin of the vibration feature processing unit 17 via a data line. A temperature sensing ring 19 is connected to the outer side of the AI ​​inference chip 15 via a data line.

[0049] Furthermore, the dynamic reconfiguration pins of the FPGA chip of the edge AI processor 2 are radially connected to four groups of DDR4 memory chips; the feature extraction component 16 is vertically inserted into the top of the AI ​​inference chip 15 via a spring pin connector;

[0050] After receiving the preprocessed data from the feature extraction component 16, the FPGA chip allocates 80% of its resources to the LSTM accelerator and 20% of its resources to CRC check through the radial pin reconfiguration logic unit. The data is then buffered in the DDR4 memory chip 145 via a micro coaxial cable and uploaded to the AI ​​inference chip 15 at a rate of 16Gbps through the PCIe slot. The LSTM operation layer calls the model parameters from the fault feature library layer, stores them in the NOR flash array, and then performs timing fault prediction: the input layer connects to the vibration feature vector and the oil feature vector, the hidden layer connects to the bidirectional LSTM network with a time step set to 10 seconds, and the output layer generates a fault probability matrix through software. The fault probability matrix includes piston ring wear, injector blockage, and bearing failure, etc.

[0051] When the temperature sensing loop detects a local temperature >85℃, it triggers active cooling of the PAL patch in the heat sink to ensure that the inference delay is stable within 50ms.

[0052] Please see Figure 1 , Figure 2 and Figure 7 One embodiment of the present invention provides: a marine internal combustion engine fault monitoring instrument based on AI technology, wherein the oil monitoring probe 9 includes an oil quality sensor 20 and an ultrasonic viscosity sensor 21. The oil quality sensor 20 is wound around the outer wall of the base tube 25, and the piezoelectric transducer ring of the ultrasonic viscosity sensor 21 is coaxially sleeved on the outer wall of the base tube 25. The data line outlet end of the piezoelectric transducer ring forms a contact. The lead wire of the oil quality sensor 20 passes through the inner hole of the base tube 25 and its end is connected to a micro socket. The micro socket and the contact are inserted into the dual-channel slot of the connector. An aviation interface group 22 is provided on the side of the outer wall of the housing 7. A satellite communicator 23 is fixedly installed on the top of the inner wall of the housing 7 through a high-frequency isolation plate. A power manager 24 is provided in the interlayer of the side of the inner wall of the housing 7.

[0053] The output of the vibration sensor array 8 is connected to the ADC interface of the feature extraction component 16 via a data cable. The oil monitoring probe 9 and the temperature sensor group 10 are connected to the AI ​​inference chip 15 via a multiplexer. The AI ​​inference chip 15 is interconnected with the storage library via a slot. The network interface of the AI ​​inference chip 15 is connected to the satellite communicator 23 via a data cable.

[0054] Furthermore, when the lubricating oil flows through the quartz flow channel, a sinusoidal voltage is applied to the spiral electrode of the oil quality sensor 20, and the real part ε' and imaginary part ε'' of the complex permittivity are measured by an impedance analyzer; when the ε'' / ε' ratio is >0.15, it is determined that there are metal particles >50μm, corresponding to bearing wear; at the same time, the piezoelectric transducer ring of the ultrasonic viscosity sensor 21 emits a 5MHz pulse, and the receiver measures the echo phase difference Δφ, and calculates the dynamic viscosity according to the formula μ=K·Δφ / f² (K is the flow channel geometric constant); the two signals The dual-channel slots, which transmit oil quality data via a micro-socket and viscosity data via a contact point, perform fusion diagnostics within the edge AI processor 2: if (ε'' / ε'>0.15) and (Δμ / μ0>0.3), where μ0 is the initial viscosity, the output is "metal wear + fuel dilution"; if (ε'' / ε'<0.05) and (Δμ / μ0>0.5), the output is "moisture contamination"; the diagnostic results are transmitted back to the remote center via satellite communicator 23, triggering the pre-delivery of spare parts.

[0055] Please see Figure 1 , Figure 3 , Figure 5 and Figure 6 The present invention provides an embodiment of a marine internal combustion engine fault monitoring instrument based on AI technology. The power manager 24 internally integrates a surge protection circuit and a battery redundancy switching module 26. The input terminal of the surge protection circuit passes through the housing 7 and connects to the ship's main power line. The output terminal of the surge protection circuit is vertically connected to the battery redundancy switching module 26 through a copper busbar. The battery redundancy switching module 26 includes a main battery compartment 27 and a backup battery compartment 28. An electromagnetic relay 29 is provided between the main battery compartment 27 and the backup battery compartment 28. The moving contact of the armature of the electromagnetic relay 29 is connected to the output bus through a phosphor bronze spring.

[0056] The shipborne monitor 3 is equipped with a vibration sensing substrate 4 and a temperature conducting copper tube 5. A piezoelectric ceramic array 6 is provided at the top of the outer wall of the vibration sensing substrate 4. The piezoelectric ceramic array 6 is embedded in the inner wall of the shipborne monitor 3 in an equilateral triangle arrangement. The temperature conducting copper tube 5 extends through the shipborne monitor 3 to the threaded mounting hole of the internal combustion engine cylinder block.

[0057] Furthermore, when the ship's power grid fluctuates, the surge protection circuit absorbs transient overvoltages >1000V and clamps the output voltage to 24±0.5V through the MOV varistor; when the main power is interrupted, the electromagnetic relay 29 switches to the backup battery compartment 28 within 15ms, that is, the armature pulls the phosphor bronze spring under the action of electromagnetic force, so that the moving contact springs from the main battery contact to the backup contact; the output bus is inserted into the double spring contact socket of the edge AI processor 2 through an L-shaped bend to ensure that the contact resistance is <10mΩ under vibration environment; sensor data is transmitted through the anti-interference bus, and the pitch of the twisted pair signal line decreases from 10cm in the low frequency band to 2cm in the high frequency band, effectively suppressing common mode interference; the drainage gap of the aluminum foil shielding layer drains condensate; the 60° braided corner armor layer can withstand a tensile force of 500N.

[0058] Please see Figure 3 , Figure 4 , Figure 5 and Figure 8 The present invention provides an embodiment of a marine internal combustion engine fault monitoring instrument based on AI technology. The system includes an AI model training server 30 and a fault decision engine 31. The AI ​​inference chip 15 and the AI ​​model training server 30 form a closed loop. The AI ​​model training server 30 is connected to the knowledge base output end through a fiber optic switch. The diagnostic results of the edge AI processor 2 are accessed to the input end of the fault decision engine 31. The output end of the fault decision engine 31 is connected to the maintenance plan generator 32. The satellite communicator 23 establishes a bidirectional link with the ship-shore communication network. The ship-shore communication network transmits data in parallel through a maritime satellite channel and a redundant 4G / 5G channel.

[0059] The AI ​​model training server 30 incorporates an LSTM optimization algorithm. The parameters output by the AI ​​model training server 30 are encrypted and compressed before being transmitted to the fault feature library. The AI ​​inference chip 15, after marking fault sample data with timestamps and watermarks, sends it back to the knowledge base. The vibration sensor array 8 collects cylinder head vibration signals, the oil monitoring probe 9 obtains the dielectric constant of the lubricating oil, the temperature sensor group 10 monitors the exhaust temperature, the vibration feature processing unit 17 performs wavelet noise reduction, the oil feature processing unit 18 performs pulse coding, and the accelerator board 14 accelerates the preprocessed data in parallel before inputting it into the AI ​​inference chip 15 via the PCIe slot. The LSTM operation layer calls the model parameters from the fault feature library and outputs a fault probability matrix. The satellite communicator 23 splits the diagnostic data packet into: core feature data transmitted in real-time via a maritime satellite channel and raw waveform data transmitted asynchronously via a 4G / 5G channel. The fault decision engine 31 aligns the core feature data and raw waveform data in time and space, and generates three levels of maintenance instructions based on the maintenance knowledge graph: Level 1 instructions are directly issued to the display terminal of the shipborne monitor 3; Level 2 instructions trigger spare parts pre-delivery; and Level 3 instructions are sent to the shore-based expert consultation center.

[0060] Furthermore, the AI ​​model training server 30 includes a fault decision engine 31 and a maintenance solution generator 32. The fault probability output by the AI ​​inference chip 15 is greater than the threshold. The feature vector sent by the maritime satellite channel includes the vibration main frequency, oil viscosity trend, etc. The data packet is marked with a timestamp watermark in the format of longitude + latitude + UTC time. The 4G / 5G channel transmits the original vibration waveform back in the port coverage area. After the remote diagnostic center receives the data, the fault decision engine 31 performs spatiotemporal alignment: first, it matches the satellite data with the asynchronous waveform according to the timestamp, then it calls the maintenance knowledge graph 208 to retrieve similar cases, and finally generates a three-level instruction.

[0061] If the remaining lifespan of the piston ring is greater than 200 hours according to the first-level command, the shipborne monitor 3 will display "Continue monitoring"; if the lifespan is 100-200 hours according to the second-level command, the spare parts will be pre-delivered to the next port; if the lifespan is less than 100 hours according to the third-level command, shore-based experts will conduct a consultation. The three-level command system reduces unnecessary downtime by 90%. At the same time, the AI ​​model training server 30 collects fault samples, updates the model parameters through the LSTM optimization algorithm, and injects them into the NOR flash memory array of the fault feature library after encryption.

[0062] Working principle: First, the vibration sensor array 8 in the multi-source sensor cluster 1, with its piezoelectric ceramic probe mounted at a 45° angle, captures the three-dimensional vibration signal of the cylinder head and transmits it via a shielded cable. The oil monitoring probe 9, through the oil quality sensor 20 and the ultrasonic viscosity sensor 21, detects the concentration and viscosity of lubricating oil metal particles, respectively, and the signals are converted and compressed by a flexible circuit. Thermocouple 12 of the temperature sensor group 10 and infrared thermometer 13 synchronously acquire the exhaust manifold temperature gradient. The piezoelectric ceramic array 6 and the temperature conduction copper pipe 5 of the onboard monitor 3 collect cylinder block vibration and temperature data. These data are transmitted to the edge AI processor 2 via a data cable.

[0063] Then, the edge AI processor 2 processes and analyzes the collected data. The vibration feature processing unit 17 of the feature extraction component 16 first uses an anti-aliasing filter to remove high-frequency noise, and then the microcontroller executes a wavelet denoising algorithm to extract specific frequency features. The oil feature processing unit 18 processes the oil data and transmits it to the AI ​​inference chip 15. The acceleration board 14 receives the preprocessed data, and the FPGA chip reconfigures the logic unit, using 80% of the resources for LSTM operation acceleration and 20% for CRC verification. After data caching, it is uploaded to the AI ​​inference chip 15 at high speed. The LSTM operation layer of the AI ​​inference chip 15 calls the model parameters of the fault feature library layer to perform time-series fault prediction on the feature vectors of vibration, oil, etc., and outputs a probability matrix containing multiple fault types. At the same time, the temperature sensing loop monitors the temperature and triggers active cooling when the temperature exceeds the limit to ensure inference efficiency.

[0064] Finally, the diagnostic results from the edge AI processor 2 are transmitted to the fault decision engine 31. If the fault probability output by the AI ​​inference chip 15 exceeds the threshold, the satellite communicator 23 splits the diagnostic data packet. Core feature data is transmitted in real time via the maritime satellite channel, while the original waveform data is transmitted asynchronously via the 4G / 5G channel. The fault decision engine 31 aligns the data in time and space and generates three-level maintenance instructions by combining the maintenance knowledge graph, thus achieving hierarchical processing. Simultaneously, the AI ​​inference chip 15 sends fault sample data back to the AI ​​model training server 30. The server updates the model parameters using the LSTM optimization algorithm, encrypts the data, and injects it into the fault feature library to complete model optimization iteration and improve the system's fault diagnosis capabilities.

[0065] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A marine internal combustion engine fault monitoring instrument based on AI technology, characterized in that: It includes a multi-source sensor cluster (1), an edge AI processor (2), and a shipborne monitor (3). The edge AI processor (2) is fixedly installed on the top of the outer wall of the multi-source sensor cluster (1), and the shipborne monitor (3) is fixedly installed on the surface of the internal combustion engine cylinder. The edge AI processor (2) is connected to the multi-source sensor cluster (1) through a data cable to establish a bidirectional data channel. The shipborne monitor (3) is provided with a vibration sensing substrate (4) and a temperature conducting copper tube (5) inside. A piezoelectric ceramic array (6) is provided at the top of the outer wall of the vibration sensing substrate (4). The piezoelectric ceramic array (6) is embedded in the inner wall of the shipborne monitor (3) in an equilateral triangle arrangement. The temperature conducting copper tube (5) extends through the shipborne monitor (3) to the threaded mounting hole of the internal combustion engine cylinder. The multi-source sensor cluster (1) is fixedly installed on the inner wall of the housing (7) by a bracket. The multi-source sensor cluster (1) is inserted into the heterogeneous interface on the outer side of the edge AI processor (2) by a cable bundle. The multi-source sensor cluster (1) includes a vibration sensor array (8), an oil monitoring probe (9), a temperature sensor group (10), and a crankshaft phase sensor (11). The vibration sensor array (8) is fixedly installed on the inner wall of the cylinder by a bracket and connected to the cylinder head bolt of the internal combustion engine. The flow channel interface of the oil monitoring probe (9) is vertically inserted into the main return oil pipe of the lubricating oil. The quartz sensing cavity of the oil monitoring probe (9) is welded to the oil pipe through a flange. The temperature sensor group (10) includes a thermocouple (12) and an infrared temperature measuring head (13). The probe end of the thermocouple (12) is nested in the internal thread of the temperature measuring hole of the exhaust manifold. The infrared temperature measuring head (13) is attached to the elbow of the exhaust pipe by a magnetic seat. The Hall chip of the crankshaft phase sensor (11) is fitted with the end face of the flywheel gear ring. The Hall chip carrier is fixed to the edge of the crankshaft by bracket bolts. The edge AI processor (2) includes an acceleration plate (14), an AI inference chip (15), and a feature extraction component (16). The acceleration plate (14) is welded to the top of the outer wall of the heat-conducting substrate through a ball grid array. The heat-conducting substrate is fixedly installed on the inner wall of the ship's control cabin. The AI ​​inference chip (15) is welded to the top of the outer wall of the acceleration plate (14), and the feature extraction component (16) is vertically inserted into the top of the outer wall of the AI ​​inference chip (15). The top of the outer wall of the accelerator board (14) is welded with a dynamic reconfiguration pin array, and the edge area is surrounded by four sets of memory chips, which are connected in series by micro data lines. The feature extraction component (16) includes two physically isolated parts, namely the vibration feature processing unit (17) in the left half and the oil feature processing unit (18) in the right half. An electromagnetic shielding plate is set between the vibration feature processing unit (17) and the oil feature processing unit (18). The vibration feature processing unit (17) is connected to the anti-aliasing filter through gold wire bonding, and the oil feature processing unit (18) is connected to the AI ​​inference chip (15) through a data cable. The AI ​​inference chip (15) is configured as a lower LSTM operation layer and an upper fault feature library layer. The fault feature library layer is bonded to the LSTM operation layer through microbumps. The LSTM operation layer is electrically connected to the PCIe slot of the acceleration board (14) through through silicon vias to form a conductive path. The flash array of the fault feature library layer is connected to the input pin of the vibration feature processing unit (17) through a data line. The AI ​​inference chip (15) has a data cable connected to the temperature sensing ring (19) through the outer wall side. The oil monitoring probe (9) includes an oil quality sensor (20) and an ultrasonic viscosity sensor (21). The oil quality sensor (20) is wound around the outer wall of the base tube (25). The piezoelectric transducer ring of the ultrasonic viscosity sensor (21) is coaxially sleeved on the outer wall of the base tube (25). The data line outlet end of the piezoelectric transducer ring forms a contact. The lead wire of the oil quality sensor (20) passes through the inner hole of the base tube (25) and its end is connected to a micro socket. The micro socket and the contact are inserted into the dual-channel slot of the connector. An aviation interface group (22) is provided on the outer side of the housing (7), a satellite communicator (23) is fixedly installed on the top of the inner wall of the housing (7) by a high frequency isolation plate, and a power manager (24) is provided in the interlayer of the inner side of the housing (7). The output of the vibration sensor array (8) is connected to the ADC interface of the feature extraction component (16) via a data cable. The oil monitoring probe (9) and the temperature sensor group (10) are connected to the AI ​​inference chip (15) via a multiplexer. The AI ​​inference chip (15) is connected to the storage library via a slot. The network interface of the AI ​​inference chip (15) is connected to the satellite communicator (23) via a data cable. The power manager (24) has a surge protection circuit and a battery redundancy switching module (26) stacked inside. The input terminal of the surge protection circuit passes through the housing (7) and is connected to the ship's main power line. The output terminal of the surge protection circuit is vertically connected to the battery redundancy switching module (26) through a copper busbar. The battery redundancy switching module (26) includes a main battery compartment (27) and a backup battery compartment (28). An electromagnetic relay (29) is set between the main battery compartment (27) and the backup battery compartment (28). The moving contact of the armature of the electromagnetic relay (29) is connected to the output bus through a phosphor bronze spring.

2. A system for monitoring marine internal combustion engine faults based on AI technology, applicable to the marine internal combustion engine fault monitoring system based on AI technology as described in claim 1, characterized in that: The system includes an AI model training server (30) and a fault decision engine (31). The AI ​​inference chip (15) and the AI ​​model training server (30) form a closed loop. The AI ​​model training server (30) is connected to the knowledge base output end through a fiber optic switch. The diagnostic results of the edge AI processor (2) are connected to the input end of the fault decision engine (31). The output end of the fault decision engine (31) is connected to the maintenance plan generator (32). The satellite communicator (23) establishes a two-way link with the ship-shore communication network. The ship-shore communication network transmits data in parallel through the maritime satellite channel and the 4G / 5G redundant channel.

3. The system for monitoring marine internal combustion engine faults based on AI technology according to claim 2, characterized in that: The AI ​​model training server (30) has a built-in LSTM optimization algorithm. The parameters output by the AI ​​model training server (30) are encrypted and compressed and then transmitted to the fault feature library. The AI ​​inference chip (15) marks the fault sample data with timestamp watermark and then sends it back to the knowledge base. The vibration sensor array (8) collects the cylinder head vibration signal, the oil monitoring probe (9) obtains the dielectric constant of the lubricating oil, the temperature sensor group (10) monitors the exhaust temperature, the vibration feature processing unit (17) performs wavelet noise reduction, the oil feature processing unit (18) performs pulse coding, and the acceleration board (14) accelerates the preprocessed data in parallel and then inputs it into the AI ​​inference chip (15) through the PCIe slot. The LSTM operation layer calls the model parameters of the fault feature library and outputs the fault probability matrix. The satellite communicator (23) splits the diagnostic data packet into: core feature data is transmitted in real time through the maritime satellite channel and raw waveform data is transmitted asynchronously through the 4G / 5G channel; the fault decision engine (31) aligns the core feature data and raw waveform data in time and space, and generates three-level maintenance instructions in combination with the maintenance knowledge graph: the first-level instructions are directly issued to the display terminal of the shipborne monitor (3); the second-level instructions trigger the pre-delivery of spare parts; and the third-level instructions are sent to the shore-based expert consultation center.

Images