Shafting fault diagnosis method and system based on ship motor howling identification

Abstract

The invention discloses a shafting fault diagnosis method based on ship motor howling identification, and the method comprises the following steps: collecting a sound signal of a ship motor and multi-source monitoring data; preprocessing the sound signals to extract time-frequency characteristics of the sound signals; inputting the time-frequency characteristics into a pre-trained recognition model to obtain a judgment result of motor squeal; probability smoothing and threshold judgment are carried out on the judgment result, and when a set condition is met, shafting diagnosis is triggered; after the shafting diagnosis is triggered, performing data fusion on the multi-source monitoring data and the time-frequency characteristics, and judging and positioning a shafting fault based on a fusion result to obtain a diagnosis result; and displaying the diagnosis result on an intelligent monitoring platform in real time, and triggering graded alarm and remote uploading. The method and the system provided by the invention have the advantages of strong real-time performance, high anti-interference capability, reliable diagnosis result, low operation and maintenance cost and the like.

Description

Technical Field

[0001] This invention relates to the field of condition monitoring and fault diagnosis technology for mechanical equipment in new energy ships, and in particular to a shafting fault diagnosis method and system based on the identification of ship motor whistling. Background Technology

[0002] As ship propulsion systems transition towards green and intelligent technologies, new energy ships are gradually adopting electric motors as the core power source for their propulsion systems. Electric motors drive propellers via shaft coupling, offering advantages such as rapid response, precise control, and relatively low noise. However, the motors and shaft systems operate under complex conditions of high power, high speed, variable load, and frequent start-stop cycles, making them prone to typical faults such as bearing wear, misalignment, shaft bending, and abnormal gear meshing. These faults not only lead to decreased propulsion efficiency and increased energy consumption but may also further trigger serious mechanical or electrical safety accidents. Particularly in highly electrified new energy ships, where propulsion motors are tightly coupled with power electronics and battery management systems, shaft system faults can easily trigger a chain reaction, threatening the stability of the entire ship's electrical system and even causing propulsion system shutdowns. Therefore, achieving real-time, intelligent fault diagnosis of the shaft systems of new energy ships is a key technology for ensuring their navigation safety and operational economy.

[0003] Currently, the fault diagnosis technologies applied to ship shafting systems mainly fall into the following categories, but each has significant limitations in the application scenarios of new energy ships:

[0004] The first category is vibration signal-based analysis methods. This method collects vibration signals using accelerometers mounted on bearing housings or engine bases and extracts fault features using time-domain, frequency-domain, or time-frequency analysis techniques. While this technology is widely used in traditional internal combustion engine ships, in the engine rooms of new energy ships, strong electromagnetic noise, power electronic switching noise, and complex structural noise severely contaminate vibration signals, leading to difficulties in feature extraction and decreased diagnostic accuracy. Furthermore, the installation location of vibration sensors is limited, and the reliability and maintenance costs of long-term operation in harsh environments with high temperature and humidity are also challenges.

[0005] The second category is monitoring methods based on speed and torque parameters. This method indirectly determines the health status by analyzing fluctuations in operating parameters such as shaft speed, torque, and power. Its advantage lies in its ability to directly utilize existing control system signals without requiring additional sensors. However, this method has low sensitivity, typically only detecting significant parameter changes in the middle to late stages of a fault, thus failing to provide early warning. Furthermore, the typical variable loads and frequent start-stop conditions of new energy vessels cause significant fluctuations in speed and torque signals, further increasing the difficulty of identifying fault characteristics from fluctuations in normal operating conditions.

[0006] The third category is offline diagnostic methods based on oil analysis: This method assesses the wear condition of bearings and gears by periodically collecting and analyzing the composition and physicochemical properties of metal abrasive particles in lubricating oil. Although effective for diagnosing wear-related faults, the entire process is time-consuming, relies on manual sampling and laboratory analysis, cannot meet the needs of real-time online monitoring, and the diagnostic results have a serious lag, making it difficult to adapt to the requirements of modern intelligent ship operation and maintenance.

[0007] The fourth category is experience-based judgment methods based on manual inspections: In some ships, the condition assessment still relies on the sensory experience of technicians through listening, touching, and observing. Although this method is direct, it is highly subjective and dependent on personnel experience. Especially in the engine room environment with complex background noise, the human ear has difficulty accurately distinguishing specific motor whistling sounds, resulting in poor reliability and the inability to achieve continuous monitoring.

[0008] The fifth category comprises offline diagnostic models based on big data and machine learning: these methods utilize historical data to train classifiers or deep learning networks for fault diagnosis. While artificial intelligence technology has demonstrated enormous potential, its successful application relies on large-scale, high-quality labeled data. In actual ship operations, data distribution is prone to drift, leading to insufficient model generalization ability. Furthermore, offline diagnostic models struggle to meet the urgent needs for real-time fault warnings and continuous condition monitoring.

[0009] In summary, existing technologies struggle to achieve real-time and accurate diagnosis of early-stage shafting faults under the complex operating conditions of new energy vessels. It is noteworthy that when shafting abnormalities occur, the motors typically emit a characteristic whistling sound, a phenomenon that offers a new direction for shafting diagnosis based on sound signals. Compared to vibration monitoring technology, sound signal acquisition is more flexible, utilizing non-contact microphone arrays; furthermore, sound signals possess rich spectral characteristics, capable of reflecting abnormal shafting conditions in the early stages of a fault. However, the noise environment in a ship's engine room is complex, encompassing propeller noise, wave noise, fan noise, and noise from other mechanical equipment, making it difficult for traditional manual listening methods to accurately identify the characteristics of motor whistling sounds. Furthermore, existing sound analysis methods largely rely on manual experience or simple spectral analysis techniques, lacking a systematic and intelligent diagnostic framework.

[0010] Therefore, there is an urgent need to propose a systematic method and device that can achieve intelligent identification based on motor whistling sound in complex noise environments and combine it with shaft fault diagnosis, so as to overcome the shortcomings of existing technologies and meet the needs of new energy ships for high safety, real-time and intelligent operation and maintenance. Summary of the Invention

[0011] In view of the above-mentioned shortcomings in current shafting fault diagnosis, this invention provides a shafting fault diagnosis method and system based on marine motor whistling identification. It can realize the intelligent identification of motor whistling and the linkage of shafting fault diagnosis in the complex noise environment of new energy ships, and has the advantages of strong real-time performance, high anti-interference ability, reliable diagnostic results, and low operation and maintenance costs.

[0012] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions:

[0013] A shafting fault diagnosis method based on marine motor whistling identification, the method comprising the following steps:

[0014] Collect sound signals from ship motors, as well as multi-source monitoring data;

[0015] The sound signal is preprocessed to extract its time-frequency features;

[0016] The time-frequency features are input into a pre-trained recognition model to obtain the judgment result of motor whistling;

[0017] The discrimination results are subjected to probability smoothing and threshold judgment. When the set conditions are met, shaft system diagnosis is triggered.

[0018] After triggering shaft system diagnosis, the multi-source monitoring data and the time-frequency characteristics are fused together, and the shaft system fault is identified and located based on the fusion result to obtain the diagnosis result;

[0019] The diagnostic results are displayed in real time on the intelligent monitoring platform, triggering tiered alarms and remote uploading.

[0020] According to one aspect of the present invention, the sound signal is acquired by a sound acquisition module deployed in the ship's engine room, the sound acquisition module including a MEMS microphone array and an analog-to-digital converter; wherein, the MEMS microphone array includes 4-16 microphones; the analog-to-digital converter has a sampling rate of not less than 44.1 kHz and a quantization accuracy of 16 bits or more.

[0021] According to one aspect of the present invention, the sound acquisition module, in conjunction with array beamforming technology, enhances the sound signal of the ship's motor, specifically as follows:

[0022] The sound signals collected by each microphone in the MEMS microphone array are weighted and superimposed using a beamforming algorithm. The output signal after processing by the beamforming algorithm satisfies the following formula:

[0023] (1)

[0024] in, For the first Signals collected by one microphone For delay compensation amount, These are the weighting coefficients.

[0025] According to one aspect of the present invention, preprocessing the sound signal to extract time-frequency features includes the following steps:

[0026] The sound signal is filtered using a bandpass filtering algorithm;

[0027] The filtered audio signal is sequentially divided into frames and windowed. Each frame is 20ms to 40ms long, the frame shift is half the frame length, and each frame signal is multiplied by a Hamming window or a Blackman window function. The windowed sound signal is obtained as shown in formula (5):

[0028] (5)

[0029] A short-time Fourier transform is performed on the windowed sound signal to obtain time-frequency characteristics, which include the frequency peak position, harmonic energy distribution, spectral centroid, and harmonic components. The transform expression is shown in formula (6):

[0030] (6)

[0031] in, For the first Frame in Complex spectrum at a frequency point For frame shift, is the length of the FFT.

[0032] According to one aspect of the present invention, the recognition model is one of a convolutional neural network, a Transformer model, or a support vector machine, used to output a discrimination result of motor whistling based on the input time-frequency features, wherein the discrimination result includes the posterior probability of the whistling category. ;

[0033] When the recognition model is a convolutional neural network, it extracts local time-frequency features through convolution operations, and its calculation satisfies formula (7):

[0034] (7)

[0035] in, Indicates the first Layer in position The output feature value, Indicates the previous level (the first) (layer) in position The input feature values, Indicates the position of the convolution kernel The weight parameters, Indicates the first Layer bias terms, Represents a nonlinear activation function;

[0036] When the recognition model is a Transformer model, it models long-term time-series dependencies through a self-attention mechanism, and its attention weights satisfy formula (8):

[0037] (8)

[0038] in, The query matrix is ​​obtained by linearly transforming the input feature vectors. The key matrix is ​​obtained by linear transformation of the input feature vectors. The value matrix is ​​obtained by linear transformation of the input eigenvectors. Query the similarity matrix between the key and the query result. The dimension of the key vector is used for scaling to avoid excessively large values. The normalization function is used to obtain the attention weights. The output matrix after attention weighting;

[0039] When the recognition model is a support vector machine, it achieves classification by constructing a hyperplane, and its discriminant function satisfies formula (9):

[0040] (9)

[0041] in, As training samples, For category labels, For Lagrange multipliers, For kernel function, This is a bias term.

[0042] According to one aspect of the present invention, the discrimination result is subjected to probability smoothing and threshold judgment, and when a set condition is met, shaft system diagnosis is triggered, specifically as follows:

[0043] When the posterior probability And the number of frames that consecutively meet the threshold condition is greater than or equal to At that time, shaft system diagnostics are triggered;

[0044] When smoothing probability Less than or equal to the release threshold And continue When the frame is reached, the trigger state is deactivated;

[0045] Among them, smoothing probability The calculation method is shown in formula (10):

[0046] (10)

[0047] in, , and For the set threshold, To smooth the window length, and These represent the consecutive frames at which the event is triggered and deactivated, respectively.

[0048] According to one aspect of the present invention, the method of identifying and locating shaft system faults based on the fusion results to obtain diagnostic results specifically includes:

[0049] The multi-source monitoring data and the time-frequency characteristics are fused together, and shaft system faults are identified by the spectrum energy ratio method. The spectrum energy ratio calculation formula is shown in formula (13):

[0050] (13)

[0051] in, Fault characteristic frequency The energy ratio at that location For the signal at frequency The power spectrum at that location, For characteristic bandwidth, and To analyze the lower and upper limits of the frequency range; when Exceeding the set threshold When this occurs, it is determined that the shaft system has a corresponding type of fault.

[0052] According to one aspect of the present invention, the graded alarm includes a first-level early warning, a second-level severe alarm, and a third-level emergency alarm, and is linked and controlled by audible and visual signal devices, event alarms, or automatic protection logic.

[0053] According to one aspect of the invention, the criterion for the graded alarm is based on the posterior probability. and energy ratio Threshold discrimination and time-domain persistence constraints, for the energy ratio Perform moving average smoothing to smooth the energy ratio. The calculation method is shown in formula (14):

[0054] (14)

[0055] in, To smooth the window length;

[0056] For tiered alarms, set smoothing energy ratios separately. threshold Posterior probability threshold The threshold satisfies and These correspond to Level 1 warning, Level 2 severe alarm, and Level 3 emergency alarm.

[0057] The criteria for the graded alarm include:

[0058] when or And continuously satisfy A level 1 warning is triggered when the frame is reached;

[0059] when or And continuously satisfy When the frame is reached, a level 2 critical alarm is triggered;

[0060] when or And continuously satisfy When the frame is reached, a Level 3 emergency alarm is triggered;

[0061] For the aforementioned graded alarms, energy ratios for hysteresis release are set respectively. threshold Posterior probability threshold and sustained frame rate ;

[0062] when And continue When the frame is reached, the corresponding alarm is deactivated; among them, satisfy , satisfy , The number of continuous frames required to clear the corresponding alarm level.

[0063] A shafting fault diagnosis system based on marine motor whistling identification, the system comprising:

[0064] Data acquisition module: Collects sound signals from the ship's motors, as well as multi-source monitoring data including vibration and rotational speed signals;

[0065] Sound signal preprocessing module: preprocesses the sound signal to extract its time-frequency features; Sound recognition module: inputs the time-frequency features into a pre-trained recognition model to obtain the discrimination result of motor whistling;

[0066] Trigger control module: Performs probability smoothing and threshold judgment on the discrimination result, and triggers shaft system diagnosis when the set conditions are met;

[0067] Shaft system diagnostic module: After triggering shaft system diagnostics, the multi-source monitoring data and the time-frequency characteristics are fused together, and the shaft system fault is identified and located based on the fusion result to obtain the diagnostic result;

[0068] Result output module: Displays the diagnostic results in real time on the intelligent monitoring platform, and triggers tiered alarms and remote uploads.

[0069] The advantages of this invention are as follows: By deploying a microphone array and combining it with beamforming technology, directional enhancement of motor whistling sounds is achieved in complex engine room noise environments, effectively suppressing interference. Furthermore, by utilizing a time-frequency feature extraction and recognition model, automated and high-precision extraction and classification of motor whistling sounds are achieved. Based on this, a trigger control mechanism based on probability smoothing and threshold judgment, along with a tiered alarm strategy with hysteresis characteristics, enables intelligent and automatic triggering of the shafting fault diagnosis process, effectively avoiding the lag caused by manual inspection and offline analysis. Simultaneously, relying on the ship intelligent monitoring platform (IAS), real-time display of diagnostic results, event recording, and remote uploading are achieved. In summary, this invention can achieve intelligent identification of motor whistling sounds and linkage with shafting fault diagnosis in complex noise environments on new energy vessels, possessing advantages such as strong real-time performance, high anti-interference capability, reliable diagnostic results, and low operation and maintenance costs. It can be widely applied to new energy ferries, fuel cell cruise ships, port operation vessels, and other green shipping equipment. Attached Figure Description

[0070] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0071] Figure 1 This is a schematic flowchart illustrating a shafting fault diagnosis method based on marine motor whistling identification as described in this invention.

[0072] Figure 2 This is a schematic diagram of the shafting fault diagnosis system based on marine motor whistling identification as described in this invention.

[0073] Figure 3 This is a schematic diagram of the convolutional neural network structure of a shafting fault diagnosis method based on marine motor whistling identification as described in this invention.

[0074] Figure 4 This is a schematic diagram of the support vector machine decision-making process of a shafting fault diagnosis method based on marine motor whistling identification as described in this invention.

[0075] Figure 5 This is a schematic diagram of the Transformer model network structure of a shafting fault diagnosis method based on marine motor whistling identification as described in this invention. Detailed Implementation

[0076] 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.

[0077] Example 1

[0078] like Figure 1 , Figure 3 , Figure 4 and Figure 5 As shown, a shafting fault diagnosis method based on marine motor whistling identification includes the following steps:

[0079] Step S1: Collect the sound signal of the ship's motor and multi-source monitoring data;

[0080] Step S2: Preprocess the sound signal to extract its time-frequency features;

[0081] Step S3: Input the time-frequency features into the pre-trained recognition model to obtain the discrimination result of motor whistling;

[0082] Step S4: Perform probability smoothing and threshold judgment on the discrimination result. When the set conditions are met, trigger shaft system diagnosis.

[0083] Step S5: After triggering shaft system diagnosis, the multi-source monitoring data and the time-frequency characteristics are fused together, and the shaft system fault is identified and located based on the fusion result to obtain the diagnosis result;

[0084] Step S6: Display the diagnostic results in real time on the intelligent monitoring platform, and trigger tiered alarms and remote uploads.

[0085] Furthermore, in step S1, collecting the ship's motor sound signal specifically includes:

[0086] High-fidelity audio signal acquisition is achieved through a sound acquisition module deployed within the ship's engine room. This module includes a MEMS microphone array and an analog-to-digital converter (ADC). The MEMS microphone array contains 4-16 microphones; the ADC has a sampling rate of at least 44.1 kHz and a quantization accuracy of 16 bits or higher. The sound acquisition module is externally protected against water, oil, and salt spray corrosion to adapt to the complex environment of the ship's engine room. The module is fixed to the side of the ship's motor using a mounting bracket made of vibration-damping material to reduce interference from mechanical vibrations.

[0087] Furthermore, the sound acquisition module, combined with array beamforming technology, enhances the sound signal of the ship's motor, specifically as follows:

[0088] Array beamforming technology uses beamforming algorithms to weightedly superimpose sound signals collected by each microphone in a MEMS microphone array. This can enhance the directivity of sound sources in a specific direction, such as the whistling sound of ship motors and their shafts, while suppressing background noise from non-target directions, such as propeller noise, wave noise, and engine room fan noise. The output signal processed by the beamforming algorithm satisfies the following formula:

[0089] (1)

[0090] in, For the first Signals collected by one microphone For delay compensation amount, These are the weighting coefficients.

[0091] Furthermore, the beamforming algorithm may employ a delay summation method, a minimum variance distortionless response (MVDR) algorithm, or an adaptive beamforming algorithm to improve the signal-to-noise ratio and achieve robust sound source enhancement.

[0092] Further, in step S1, multi-source monitoring data is collected, including vibration signals, rotational speed signals, and propulsion system operating data, specifically:

[0093] Multiphysics signals are synchronously acquired through a sensor network installed at key locations in the ship's propulsion shafting. Vibration signals are collected by sensors positioned on the main shaft bearings, intermediate bearings, and thrust bearing housings, measuring both radial and axial directions. Rotational speed signals are obtained via magnetoelectric or photoelectric encoders mounted at the free ends of the shafting. Propulsion system operating data, including key parameters such as the output power of the main engine and propulsion motors, and shaft torque, are acquired in real-time through an Intelligent Detection System (IAS) platform or directly from devices such as power meters and torque meters via data buses (e.g., CAN, Modbus).

[0094] Further, in step S2, the sound signal is preprocessed to extract its time-frequency features; specifically, this includes the following steps:

[0095] Step S21: The sound signal is filtered using a bandpass filtering algorithm; specifically:

[0096] The acquired sound signal is filtered using a bandpass filter algorithm to remove low-frequency mechanical vibrations and high-frequency electromagnetic interference. Its typical frequency range is 200Hz to 8kHz. The time-domain expression of the bandpass filter is shown in formula (2).

[0097] (2)

[0098] in, This is the filtered output signal. The input is the original sound signal. The unit impulse response of the bandpass filter. This represents the filter order.

[0099] The ideal transfer function of this bandpass filter in the frequency domain can be expressed as:

[0100] (3)

[0101] in, Let be the frequency response function of the bandpass filter. The lower limit frequency of the bandpass filter (preferably 200Hz) is used. To determine the upper limit frequency of the bandpass filter (preferably 8000Hz), the bandpass filter can be designed using a finite impulse response (FIR) filter, whose impulse response is:

[0102] (4)

[0103] in, This represents the impulse response of the low-pass filter at the corresponding cutoff frequency.

[0104] Step S22 involves sequentially performing frame segmentation and windowing operations on the filtered audio signal; specifically:

[0105] The filtered signal is sequentially processed through frame segmentation and windowing to ensure temporal continuity and suppress spectral leakage. Frame segmentation is performed on the filtered signal, with each frame lasting 20–40 ms and the frame shift set to half the frame length to maintain temporal continuity. Before entering the frequency domain transformation, a Hamming window function or a Blackman window function is applied to each frame. The windowed signal is obtained as follows:

[0106] (5)

[0107] in, This represents the signal after framing. It is either a Hamming window function or a Blackman window function.

[0108] Step S23: Perform a short-time Fourier transform on the windowed audio signal to obtain its time-frequency characteristics. Specifically:

[0109] After framing and windowing, a Short-Time Fourier Transform (STFT) is performed on each frame of signal to obtain the time-varying spectral distribution. The STFT can extract the time-frequency characteristics of the ship's motor whistling sound, including peak frequency location, harmonic energy distribution, spectral centroid, and harmonic components. Its transformation expression is as follows:

[0110] (6)

[0111] in, For the first Frame in Complex spectrum at a frequency point For frame shift, The length of the Fast Fourier Transform (FFT) is given. Using the aforementioned short-time Fourier transform, the frequency peaks, energy distribution, spectral centroid, and harmonic components of the ship's motor whistling sound can be extracted to characterize the abnormal state of the motor shaft system. These time-frequency characteristics reflect the dynamic properties of the motor shaft system under abnormal conditions; especially when the motor experiences faults such as bearing defects, misalignment, or abnormal gear meshing, the spectrum obtained based on the short-time Fourier transform will exhibit typical modulation side frequencies and energy concentration phenomena.

[0112] Further, in step S3, the time-frequency features are input into the pre-trained recognition model to obtain the discrimination result of the motor whistling; specifically:

[0113] The extracted time-frequency features are input into a pre-trained sound recognition model, which outputs a discrimination result for "whistling" or "non-whistling". The sound recognition model is one of a Convolutional Neural Network (CNN), a Transformer model, or a Support Vector Machine (SVM). Each model takes the time-frequency features obtained through short-time Fourier transform preprocessing as input and outputs a classification result to determine whether there is a ship motor whistling sound, including the posterior probability of the "whistling" category. The specific implementation is as follows:

[0114] When the recognition model is a convolutional neural network, such as Figure 2 As shown, Convolutional Neural Networks (CNNs) extract local time-frequency features through convolution operations. Layer in position The output eigenvalues ​​satisfy:

[0115] (7)

[0116] in, Indicates the first Layer in position The output feature value; Indicates the previous level (the first) (layer) in position The input feature values; Indicates the position of the convolution kernel Weight parameters; Indicates the first Layer bias terms; This represents a non-linear activation function, such as the ReLU function or the Sigmoid function.

[0117] When the recognition model is a Transformer model, such as Figure 4 As shown, the ransformer model models long-term temporal dependencies through a self-attention mechanism, and its attention weight calculation satisfies:

[0118] (8)

[0119] in, The query matrix is ​​obtained by linear transformation of the input feature vectors. The key matrix is ​​obtained by linear transformation of the input feature vectors; The value matrix is ​​obtained by linear transformation of the input feature vectors; Query the similarity matrix between the query and the key; The dimension of the key vector is used for scaling to avoid excessively large values; The normalization function is used to generate attention weights; The output matrix after attention weighting.

[0120] When the recognition model is a support vector machine, such as Figure 3 As shown, Support Vector Machines (SVMs) achieve classification by constructing a hyperplane, and their discriminant function satisfies:

[0121] (9)

[0122] in, As training samples, For category labels, For Lagrange multipliers, For kernel function, This is a bias term.

[0123] Furthermore, in step S4, the discrimination result is subjected to probability smoothing and threshold judgment. When a set condition is met, shaft system diagnosis is triggered; specifically:

[0124] When the sound recognition model detects a whistling sound from the ship's motors, the shafting diagnostic process is initiated, which includes the following operations:

[0125] The posterior probability of the "howling" category output by the sound recognition model Perform threshold discrimination:

[0126] When the posterior probability of the howling category And the number of frames that consecutively meet the threshold condition is greater than or equal to At that time, shaft system diagnostics are triggered;

[0127] When the smoothed probability Less than or equal to the release threshold And continue When the frame is reached, the trigger state is deactivated.

[0128] Among them, smoothing probability The calculation method is as follows:

[0129] (10)

[0130] in , and The set threshold; To smooth the window length; and These represent the consecutive frame counts for triggering and deactivation, respectively. The trigger axis diagnostic process further incorporates a hysteresis control mechanism to reduce false triggering caused by background noise or transient interference.

[0131] Further, in step S5, after triggering the shaft system diagnosis, the multi-source monitoring data and the time-frequency characteristics are fused, and the shaft system fault is identified and located based on the fusion result to obtain a diagnosis result; specifically:

[0132] Step S51: Perform data fusion between the multi-source monitoring data and the time-frequency features;

[0133] To improve the robustness and accuracy of shaft system fault diagnosis, vibration signals will be included. Speed ​​signal With propulsion system operating data (such as power) Torque Multi-source monitoring data is fused with time-frequency characteristics, and the signals from each source are aligned to a time axis based on the sound frame. (frame shift) ):

[0134] (11)

[0135] Features are extracted from each source signal, and the extracted features are normalized to zero mean and unit variance; the feature vectors of each source are represented as follows:

[0136] (12a)

[0137] The normalization formula is:

[0138] (12b)

[0139] Joint features are formed by feature concatenation, or fusion indices are obtained through linear weighting; where:

[0140] The joint features formed by feature concatenation are:

[0141] (13a)

[0142] The fusion index obtained by linear weighting is:

[0143] (13b)

[0144] in, Possible posterior probabilities based on the sound side. ; The vibrational energy ratio or the envelope spectrum amplitude can be used; and This is a normalized index based on rotational speed and propulsion conditions.

[0145] Step S52: Determine shaft system faults using the spectral energy ratio method; specifically:

[0146] Based on the energy ratio criterion method, determine and locate bearing wear, alignment deviation, shaft bending, or abnormal gear meshing in the ship's shafting system;

[0147] The spectral energy ratio method is used for fault diagnosis of ship shafting, and its calculation formula is as follows:

[0148] (13)

[0149] in, Fault characteristic frequency The energy ratio at the location; For the signal at frequency Power spectrum at; The characteristic bandwidth; and To analyze the lower and upper limits of the frequency range; when Exceeding the set threshold When this is determined, the shaft system is found to have a corresponding type of fault, such as bearing wear, misalignment, shaft bending, or abnormal gear meshing.

[0150] In addition, methods such as envelope analysis and order analysis can be combined to further improve the accuracy of identifying complex shaft system faults.

[0151] Furthermore, in step S6, the diagnostic results are displayed in real time on the intelligent monitoring platform, and a tiered alarm and remote upload are triggered. Specifically:

[0152] The Intelligent Monitoring System (IAS) platform displays real-time diagnostic results for the ship's shafting system and triggers tiered alarms based on these results. These tiered alarms include three levels: Level 1 warning, Level 2 severe alarm, and Level 3 emergency alarm, which are linked and controlled via audible and visual signals, event alarms, or automatic protection logic. Simultaneously, it generates timestamped fault event records and uploads them to the shore-based monitoring center via a communication link to enable remote operation and maintenance, historical data tracing, and fault statistical analysis.

[0153] The criterion for the graded alarm is based on the posterior probability. and energy ratio Threshold discrimination and time-domain persistence constraints, for and The indicator is smoothed by a moving average, and the smoothing probability is... The calculation method is shown in formula (10). The calculation method is shown in formula (9):

[0154] (14)

[0155] in, To smooth the window length;

[0156] For tiered alarms, set smoothing energy ratios separately. threshold Posterior probability threshold The threshold satisfies and These correspond to Level 1 warning, Level 2 severe alarm, and Level 3 emergency alarm.

[0157] The criteria for the graded alarm include:

[0158] when or And continuously satisfy A level 1 warning is triggered when the frame is reached;

[0159] when or And continuously satisfy When the frame is reached, a level 2 critical alarm is triggered;

[0160] when or And continuously satisfy When the frame is reached, a level three emergency alarm is triggered; and the IAS safety management module is activated to execute power reduction or shutdown protection logic.

[0161] To reduce false triggering, a hysteresis release threshold is adopted. , ;

[0162] For the aforementioned graded alarms, energy ratios for hysteresis release are set respectively. threshold Posterior probability threshold and sustained frame rate ;

[0163] when And continue When the frame is reached, the corresponding alarm is deactivated.

[0164] in, satisfy , satisfy , The number of continuous frames required to clear the corresponding alarm level.

[0165] Furthermore, the diagnostic result trigger also supports comprehensive triggering based on fusion indicators, which are calculated as follows:

[0166] (15)

[0167] when And continue When the frame is reached, a level 2 or 3 alarm is triggered directly; the above parameters The determination is based on the ship type, engine room sound field, and equipment level.

[0168] Example 2

[0169] like Figure 2 As shown, a shafting fault diagnosis system based on marine motor whistling identification is disclosed. The system includes:

[0170] Data acquisition module M1: Acquires sound signals from the ship's motors, as well as multi-source monitoring data;

[0171] Audio signal preprocessing module M2: preprocesses the audio signal to extract its time-frequency features;

[0172] Sound recognition module M3: Inputs the time-frequency features into the pre-trained recognition model to obtain the judgment result of motor whistling;

[0173] Trigger control module M4: Performs probability smoothing and threshold judgment on the discrimination result, and triggers shaft system diagnosis when the set conditions are met;

[0174] Shaft system diagnostic module M5: After triggering shaft system diagnostics, it fuses the multi-source monitoring data with the time-frequency characteristics, and judges and locates shaft system faults based on the fusion results to obtain diagnostic results;

[0175] Result output module M6: Displays the diagnostic results in real time on the intelligent monitoring platform, and triggers tiered alarms and remote uploads.

[0176] The advantages of this invention are as follows: By deploying a microphone array and combining it with beamforming technology, directional enhancement of motor whistling sounds is achieved in complex engine room noise environments, effectively suppressing interference. Furthermore, by utilizing a time-frequency feature extraction and recognition model, automated and high-precision extraction and classification of motor whistling sounds are achieved. Based on this, a trigger control mechanism based on probability smoothing and threshold judgment, along with a tiered alarm strategy with hysteresis characteristics, enables intelligent and automatic triggering of the shafting fault diagnosis process, effectively avoiding the lag caused by manual inspection and offline analysis. Simultaneously, relying on the ship intelligent monitoring platform (IAS), real-time display of diagnostic results, event recording, and remote uploading are achieved. In summary, this invention can achieve intelligent identification of motor whistling sounds and linkage with shafting fault diagnosis in complex noise environments on new energy vessels, possessing advantages such as strong real-time performance, high anti-interference capability, reliable diagnostic results, and low operation and maintenance costs. It can be widely applied to new energy ferries, fuel cell cruise ships, port operation vessels, and other green shipping equipment.

[0177] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A shafting fault diagnosis method based on marine motor whistling identification, characterized in that, The method includes the following steps: Collect sound signals from ship motors, as well as multi-source monitoring data; The sound signal is preprocessed to extract its time-frequency features; The time-frequency features are input into a pre-trained recognition model to obtain the judgment result of motor whistling; The discrimination results are subjected to probability smoothing and threshold judgment. When the set conditions are met, shaft system diagnosis is triggered. After triggering shaft system diagnosis, the multi-source monitoring data and the time-frequency characteristics are fused together, and the shaft system fault is identified and located based on the fusion result to obtain the diagnosis result; The diagnostic results are displayed in real time on the intelligent monitoring platform, triggering tiered alarms and remote uploading.

2. The shaft system fault diagnosis method according to claim 1, characterized in that, The sound signal is acquired by a sound acquisition module installed in the ship's engine room. The sound acquisition module includes a MEMS microphone array and an analog-to-digital converter. The MEMS microphone array contains 4-16 microphones. The analog-to-digital converter has a sampling rate of not less than 44.1 kHz and a quantization accuracy of 16 bits or more.

3. The shaft system fault diagnosis method according to claim 2, characterized in that, The sound acquisition module, combined with array beamforming technology, enhances the sound signal of the ship's motor, specifically as follows: The sound signals collected by each microphone in the MEMS microphone array are weighted and superimposed using a beamforming algorithm. The output signal after processing by the beamforming algorithm satisfies the following formula: （1） in, For the first Signals collected by one microphone For delay compensation amount, These are the weighting coefficients.

4. The shaft system fault diagnosis method according to claim 1, characterized in that, Preprocessing the audio signal to extract time-frequency features includes the following steps: The sound signal is filtered using a bandpass filtering algorithm; The filtered audio signal is sequentially divided into frames and windowed. Each frame is 20ms to 40ms long, the frame shift is half the frame length, and each frame signal is multiplied by a Hamming window or a Blackman window function. The windowed sound signal is obtained as shown in formula (5): （5） A short-time Fourier transform is performed on the windowed sound signal to obtain time-frequency characteristics, which include the frequency peak position, harmonic energy distribution, spectral centroid, and harmonic components. The transform expression is shown in formula (6): （6） in, For the first Frame in Complex spectrum at a frequency point For frame shift, is the length of the FFT.

5. The shaft system fault diagnosis method according to claim 1, characterized in that, The recognition model is one of a convolutional neural network, a Transformer model, or a support vector machine, used to output a discrimination result of motor whistling based on the input time-frequency features, wherein the discrimination result includes the posterior probability of the whistling category. ; When the recognition model is a convolutional neural network, it extracts local time-frequency features through convolution operations, and its calculation satisfies formula (7): （7） in, Indicates the first Layer in position The output feature value, Indicates the previous level (the first) (layer) in position The input feature values, Indicates the position of the convolution kernel The weight parameters, Indicates the first Layer bias terms, Represents a nonlinear activation function; When the recognition model is a Transformer model, it models long-term time-series dependencies through a self-attention mechanism, and its attention weights satisfy formula (8): （8） in, The query matrix is ​​obtained by linearly transforming the input feature vectors. The key matrix is ​​obtained by linear transformation of the input feature vectors. The value matrix is ​​obtained by linear transformation of the input eigenvectors. Query the similarity matrix between the key and the query result. The dimension of the key vector is used for scaling to avoid excessively large values. The normalization function is used to obtain the attention weights. The output matrix after attention weighting; When the recognition model is a support vector machine, it achieves classification by constructing a hyperplane, and its discriminant function satisfies formula (9): （9） in, As training samples, For category labels, For Lagrange multipliers, For kernel function, This is a bias term.

6. The shaft system fault diagnosis method according to claim 1, characterized in that, The discrimination results are subjected to probability smoothing and threshold judgment. When the set conditions are met, shaft system diagnosis is triggered, specifically: When the posterior probability And the number of consecutive frames that meet the threshold condition is greater than or equal to At that time, shaft system diagnosis is triggered; When smoothing probability Less than or equal to the release threshold And continue When the frame is reached, the trigger state is deactivated; Among them, smoothing probability The calculation method is shown in formula (10): （10） in, , and For the set threshold, To smooth the window length, and These represent the consecutive frames at which the event is triggered and deactivated, respectively.

7. The shaft system fault diagnosis method according to claim 6, characterized in that, The method of identifying and locating shaft system faults based on the fusion results to obtain diagnostic results is as follows: The multi-source monitoring data and the time-frequency characteristics are fused together, and shaft system faults are identified by the spectrum energy ratio method. The spectrum energy ratio calculation formula is shown in formula (13): （13） in, Fault characteristic frequency The energy ratio at that location For the signal at frequency The power spectrum at that location, For characteristic bandwidth, and To analyze the lower and upper limits of the frequency range; when Exceeding the set threshold When this occurs, it is determined that the shaft system has a corresponding type of fault.

8. The shaft system fault diagnosis method according to claim 7, characterized in that, The tiered alarm system includes Level 1 early warning, Level 2 severe alarm, and Level 3 emergency alarm, and is linked and controlled through audible and visual signal devices, event alarms, or automatic protection logic.

9. The shaft system fault diagnosis method according to claim 8, characterized in that, The criterion for the graded alarm is based on the posterior probability. and energy ratio Threshold discrimination and time-domain persistence constraints, for the energy ratio Perform moving average smoothing to smooth the energy ratio. The calculation method is shown in formula (14): （14） in, To smooth the window length; For tiered alarms, set smoothing energy ratios separately. threshold Posterior probability threshold The threshold satisfies and These correspond to Level 1 warning, Level 2 severe alarm, and Level 3 emergency alarm. The criteria for the graded alarm include: when or And continuously satisfy A level 1 warning is triggered when the frame is reached; when or And continuously satisfy When the frame is reached, a level 2 critical alarm is triggered; when or And continuously satisfy When the frame is reached, a Level 3 emergency alarm is triggered; For the aforementioned graded alarms, energy ratios for hysteresis release are set respectively. threshold Posterior probability threshold and sustained frame rate ; when And continue When the frame is reached, the corresponding alarm is deactivated; among them, satisfy , satisfy , The number of continuous frames required to clear the corresponding alarm level.

10. A shafting fault diagnosis system based on marine motor whistling identification, characterized in that, The system includes: Data acquisition module: Collects sound signals from the ship's motors, as well as multi-source monitoring data including vibration and rotational speed signals; Sound signal preprocessing module: preprocesses the sound signal to extract its time-frequency features; Sound recognition module: inputs the time-frequency features into a pre-trained recognition model to obtain the discrimination result of motor whistling; Trigger control module: Performs probability smoothing and threshold judgment on the discrimination result, and triggers shaft system diagnosis when the set conditions are met; Shaft system diagnostic module: After triggering shaft system diagnostics, the multi-source monitoring data and the time-frequency characteristics are fused together, and the shaft system fault is identified and located based on the fusion result to obtain the diagnostic result; Result output module: Displays the diagnostic results in real time on the intelligent monitoring platform, and triggers tiered alarms and remote uploads.

Images