Vibration isolation state detection equipment for marine diesel generating set

Abstract

The invention belongs to the technical field of vibration isolation detection, and particularly relates to a ship diesel generating set vibration isolation state detection device which collects multi-dimensional parameter sequences of set body vibration, basic vibration and vibration isolation system temperature field distribution through a sensor array. A first characteristic value representing spectrum energy distribution, a second characteristic value representing time domain statistical characteristics and a third characteristic value representing temperature field uniformity and hot spot intensity are calculated respectively, and the comprehensive module carries out weighted fusion on the three characteristic values to generate a real-time risk index; the twinning response module triggers a corresponding adjustment action based on the index and performs simulation optimization in combination with a digital twinning model; accurate diagnosis and predictive protection of faults such as rotor unbalance, vibration isolator impact, bolt looseness and abnormal temperature rise are achieved, and the operation reliability and safety of the ship generator set are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of vibration isolation testing technology, and particularly relates to a vibration isolation status testing device for marine diesel generator sets. Background Technology

[0002] In existing technologies, the condition monitoring of vibration isolation systems for marine diesel generator sets mainly relies on threshold alarm mechanisms based on single physical parameters, such as monitoring vibration amplitude or temperature values ​​individually. An alarm is triggered when the detected value exceeds a preset fixed threshold. However, marine diesel generator sets operate in a complex multi-physics coupled environment, and faults in their vibration isolation systems often manifest as a correlation and co-evolution of vibration and temperature characteristics. Single-parameter monitoring methods cannot effectively identify complex faults caused by multiple factors, such as rotor imbalance, misalignment, isolator impact, loose bolts, and overload aging. The limitations of this monitoring method make it difficult for the system to accurately distinguish fault types and locate fault positions. Furthermore, it fails to capture subtle changes in characteristics in the early stages of a fault, frequently resulting in false alarms, missed alarms, or alarm lag, severely impacting the safe and stable operation of marine generator sets. Therefore, there is an urgent need for a vibration isolation condition detection scheme that can integrate multi-physics information and achieve accurate early diagnosis. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes a vibration isolation status detection device for marine diesel generator sets. This device collects multidimensional parameter sequences of generator set vibration, foundation vibration, and temperature field distribution of the vibration isolation system through a sensor array. It calculates a first characteristic value representing the spectral energy distribution, a second characteristic value representing the time-domain statistical characteristics, and a third characteristic value representing the temperature field uniformity and hot spot intensity. A comprehensive module performs weighted fusion of the three characteristic values ​​to generate a real-time risk index. A twin response module triggers corresponding adjustment actions based on this index and performs simulation optimization in conjunction with a digital twin model. This invention achieves accurate diagnosis and predictive protection against faults such as rotor imbalance, vibration isolator impact, bolt loosening, and abnormal temperature rise, significantly improving the operational reliability and safety of marine generator sets.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] The vibration isolation status detection equipment for marine diesel generator sets includes: a data acquisition module, a spectrum analysis module, a statistical analysis module, a temperature field module, a comprehensive module, and a twin response module.

[0006] The acquisition module uses a configured sensor array combined with a filter to perform signal preprocessing and acquire the first real-time parameter sequence, the second real-time parameter sequence, and the third real-time parameter sequence of the ship's diesel generator set.

[0007] The spectrum analysis module calculates a first characteristic value characterizing the energy distribution of the body's vibration spectrum based on the first real-time parameter sequence.

[0008] The statistical analysis module calculates a second eigenvalue characterizing the time-domain statistical properties of the foundation vibration based on the second real-time parameter sequence.

[0009] The temperature field module calculates a third characteristic value characterizing the uniformity of the temperature field distribution and the intensity of hot spots based on the third real-time parameter sequence.

[0010] The comprehensive module performs a weighted combination based on the first feature value, the second feature value, and the third feature value to generate a real-time risk indicator that characterizes the current vibration isolation status and health risk level.

[0011] The twin response module is used to respond to the real-time risk indicator meeting the preset trigger condition by selecting the adjustment action corresponding to the trigger condition from an action set including at least three adjustment actions and combining it with the twin simulation algorithm to perform real-time simulation and real-time risk indicator monitoring and adjustment.

[0012] Specifically, the first real-time parameter sequence characterizes the vibration state of the ship's generator set, the second real-time parameter sequence characterizes the vibration state of the generator set's foundation or vibration isolation system, and the third real-time parameter sequence characterizes the temperature field distribution in key areas of the vibration isolation system; the first characteristic value is used to indicate the risk of rotor imbalance or misalignment during generator set operation; the time-domain statistical characteristics are related to the random interference between the generator set vibration and the external environment vibration, and are used to indicate the risk of vibration isolator impact or bolt loosening; the third characteristic value is used to indicate the risk of abnormal temperature rise caused by at least one of the following abnormalities: abnormal overload, abnormal aging, or abnormal internal friction of the vibration isolator; the adjustment actions include at least issuing a warning signal, issuing an alarm signal, performing a generator set load reduction operation, or performing a generator set switching operation.

[0013] Specifically, signal preprocessing is performed using a configured sensor array combined with filters, including:

[0014] The original vibration signals and original infrared images of the ship's diesel generator set are simultaneously acquired through a configured sensor array.

[0015] Based on the original vibration signal, impedance matching and signal amplification are performed through a charge amplifier, and synchronous sampling is performed using a 16-bit precision analog-to-digital converter at a sampling rate no less than a preset multiple of the highest analysis frequency to obtain a discrete digital vibration sequence.

[0016] Based on the discretized digital vibration sequence, a segment of the signal under stationary conditions of the equipment is extracted or the stopband characteristics of a high-pass filter are used to calculate the DC component of the discretized digital vibration sequence.

[0017] The DC component of the discretized digital vibration sequence is subtracted from each data point in the discretized digital vibration sequence to obtain a vibration signal with zero bias.

[0018] Specifically, signal preprocessing using a configured sensor array combined with filters also includes:

[0019] The zero-bias vibration signal is input into a fourth-order Butterworth low-pass filter with a cutoff frequency set to a preset multiple of the unit's highest operating frequency to filter out high-frequency vibration signals above the unit's highest operating frequency, thereby obtaining the first pre-processed vibration signal.

[0020] The first preprocessed vibration signal is input into a linear phase FIR bandpass filter that includes the passband frequency range constructed from the fundamental frequency corresponding to the rated speed of the unit and the Hanning window function to obtain the first real-time parameter sequence and the second real-time parameter sequence.

[0021] Based on the reference data calibrated in advance at different temperature points using a standard reference blackbody, a two-point correction calculation is performed on each frame of the input raw infrared image to obtain the initial calibrated infrared image;

[0022] Based on the initial calibrated infrared image, a two-dimensional convolution operation is performed using a 3×3 pixel Gaussian convolution kernel to obtain a spatially smooth temperature field image.

[0023] The temporal median filter is applied independently to each pixel location in the spatially smoothed temperature field image to obtain the preprocessed third real-time parameter sequence.

[0024] Specifically, the calculation of the first eigenvalue characterizing the energy distribution of the body's vibration spectrum includes:

[0025] Based on the first real-time parameter sequence, the time-domain vibration signal is converted into a frequency-domain spectrum using the fast Fourier transform algorithm to obtain complex spectrum data containing amplitude and phase information;

[0026] Based on the complex spectrum data and the preset frequency band feature division rules, a set of frequency band intervals is obtained; the set of frequency band intervals includes the fundamental frequency band, the second octave band, and the high-frequency feature band.

[0027] Integrate the squared amplitude of each frequency point in each frequency band of the frequency band set to obtain a frequency band energy value sequence; the frequency band energy value sequence includes fundamental band energy value, second harmonic band energy value and high-frequency characteristic band energy value.

[0028] Based on the weighted algorithm combining the fundamental band energy value, the second harmonic band energy value, and the high-frequency characteristic band energy value, a first characteristic value characterizing the energy distribution of the body's vibration spectrum is obtained.

[0029] Specifically, the calculation of the second eigenvalue, which characterizes the time-domain statistical properties of foundation vibration, includes:

[0030] Based on the second real-time parameter sequence combined with the configured fixed-duration sliding time window, multiple overlapping sub-time period signal sets are obtained;

[0031] Based on the signal set of each sub-time period, the time-domain statistical parameter set of each sub-time period signal is calculated in parallel: the time-domain statistical parameter set includes the peak-to-peak value, kurtosis value, peak factor and impulse index of the sub-time period signal;

[0032] Based on the preset interference component threshold rules, the time-domain statistical parameter set is filtered to obtain the filtered effective time-domain statistical parameter set.

[0033] Based on the filtered effective time-domain statistical parameter set combined with the preset weighted contribution weight, the feature value of each effective sub-time period is obtained.

[0034] Based on the characteristic values ​​of all valid sub-time periods, the 95th percentile is taken as the second characteristic value characterizing the time-domain statistical properties of the foundation vibration.

[0035] Specifically, the calculation of the third eigenvalue, which characterizes the uniformity of the temperature field distribution and the intensity of hot spots, includes:

[0036] Based on the third real-time parameter sequence, an adaptive threshold segmentation algorithm is used to extract hotspot candidate regions in the temperature field image; the adaptive threshold is determined by calculating the inter-peak threshold of the image grayscale histogram.

[0037] Perform morphological closing operations on the hotspot candidate regions to obtain a set of effective hotspot regions;

[0038] Based on the set of effective hotspot regions, a set of hotspot feature parameters is calculated; the set of hotspot feature parameters includes the percentage of the total area of ​​hotspot regions and the difference between the highest temperature of the hotspot region and the average temperature of the entire region.

[0039] Based on the pixel temperature values ​​of the temperature field image, the coefficient of variation of the global temperature distribution is calculated to characterize the uniformity.

[0040] By combining preset uniformity weights and hotspot intensity weights, the coefficient of variation and hotspot characteristic parameter set are weighted and fused to obtain a third characteristic value that characterizes the uniformity of the temperature field distribution and the intensity of the hotspots.

[0041] Specifically, real-time simulation and real-time risk indicator monitoring and adjustment are performed by combining twin simulation algorithms, including:

[0042] The real-time risk index value, which represents the current vibration isolation status and health risk level, is compared with the preset risk index value threshold to make an overall anomaly judgment. When the real-time risk index value is greater than the preset risk index value threshold, the vibration isolation status is determined to be globally abnormal, an alarm is triggered, and the degree of vibration isolation status abnormality value is obtained based on the real-time risk index value.

[0043] When an anomaly is determined, the anomaly degree corresponding to the current global anomaly type combination set is obtained by combining the vibration isolation state anomaly degree value, the first feature value, the second feature value, the third feature value, and the preset feature value-fault type-anomaly degree mapping table.

[0044] Based on the anomaly severity corresponding to the current global anomaly type combination set, combined with the anomaly type-adjustment action type-adjustment severity mapping table, a global candidate adjustment action parameter set is obtained; the adjustment severity includes the adjustment priority of each adjustment action, the adjustment time length of the corresponding adjustment action, and the size of the adjustment unit range.

[0045] Specifically, combining twin simulation algorithms for real-time simulation and real-time risk indicator monitoring and adjustment also includes:

[0046] Based on the global candidate adjustment action parameter set combined with the twin simulation model constructed by the ship diesel generator set, after abnormal simulation adjustment, the global inference response dataset is obtained.

[0047] Based on the global simulation response dataset and combined with a multi-objective optimization decision-making algorithm, the optimal set of adjustment actions is obtained with the goal of maximizing the preset comprehensive decision factor. The performance evaluation indicators include the change in vibration transmissibility of the vibration isolation system, the magnitude of hot spot temperature drop, and the system stability recovery time. The comprehensive decision factor is obtained by weighting the comprehensive impact score and the comprehensive performance score. The comprehensive performance score is obtained by weighting the vibration suppression rate, temperature drop magnitude, and state stability recovery time under real-time monitoring. The comprehensive impact score is obtained by weighting the execution time of the optimal set of adjustment actions, equipment wear and tear, and the probability of action execution risk.

[0048] Specifically, combining twin simulation algorithms for real-time simulation and real-time risk indicator monitoring and adjustment also includes:

[0049] When the real-time risk indicator value is less than or equal to the preset risk indicator value threshold and at least one of the first feature value, the second feature value, and the third feature value is greater than the preset abnormality judgment threshold, the vibration isolation state is judged to be locally abnormal and an alarm is triggered.

[0050] Based on the feature value set corresponding to the local anomaly in the vibration isolation state, the feature value magnitude is combined with the feature value-fault type-anomaly degree mapping table to obtain the anomaly degree corresponding to the current local anomaly type combination set.

[0051] Based on the anomaly degree corresponding to the current set of local anomaly type combinations, combined with the anomaly type-adjustment action type-adjustment degree mapping table, a set of local candidate adjustment action parameters is obtained;

[0052] Based on the local candidate adjustment action parameter set, the local optimal adjustment action combination set is obtained by repeating the twin simulation and multi-objective optimization process.

[0053] When the real-time risk indicator value is less than or equal to the preset risk indicator value threshold and the first feature value, the second feature value, and the third feature value are all less than the preset anomaly judgment feature, the changing trend of the first feature value, the second feature value, and the third feature value is monitored in real time according to the preset data interval. When the difference between at least one feature value and the corresponding preset anomaly judgment threshold is less than the preset risk warning threshold, an early warning operation is performed. At the same time, the warning level is determined according to the changing rate of the corresponding feature value, and a real-time warning of the corresponding level is performed.

[0054] Compared with the prior art, the beneficial effects of the present invention are:

[0055] This invention addresses the shortcomings of existing technologies by employing multi-sensor data fusion and an intelligent decision-making mechanism to achieve comprehensive monitoring and precise protection of the vibration isolation status of marine diesel generator sets. In particular, it constructs a multi-dimensional feature extraction system encompassing vibration spectrum, time-domain statistics, and temperature field distribution, effectively overcoming the insensitivity of traditional single-parameter monitoring to complex faults. A multi-physics coupling model is built using digital twin technology, and simulation analysis enables pre-verification of adjustment actions and parameter optimization. A hierarchical early warning and multi-objective decision-making mechanism is established, achieving early warning through eigenvalue trend analysis and precise selection of adjustment actions based on comprehensive decision factors. This system significantly improves the timeliness of fault detection, the accuracy of diagnosis, and the effectiveness of protection, providing a reliable guarantee for the safe operation of marine power systems. Attached Figure Description

[0056] Figure 1 This is a module diagram of the vibration isolation status detection equipment for marine diesel generator sets according to Embodiment 1 of the present invention;

[0057] Figure 2 This is a flowchart of the preprocessing process in Embodiment 1 of the present invention;

[0058] Figure 3 This is a flowchart of spectrum analysis in Embodiment 1 of the present invention. Detailed Implementation

[0059] Example 1

[0060] This application applies to industrial equipment condition monitoring scenarios that require complex multi-physics coupled monitoring and intelligent maintenance decision-making for the health management of marine power systems. In particular, it relates to a vibration isolation condition detection system for marine diesel generator sets with multi-source sensor data fusion, dynamic risk prediction, and adaptive protection control functions. In the process of monitoring the condition of such equipment, due to the time-varying nature of the unit's operating conditions, the complexity of environmental interference, and the nonlinear characteristics of fault evolution, traditional monitoring systems are prone to false alarms, missed alarms, or delayed responses when lacking multi-feature fusion and predictive maintenance mechanisms.

[0061] Typical technical challenges in monitoring the vibration isolation status of marine diesel generator sets include, but are not limited to:

[0062] In the continuous monitoring of the health status of vibration isolation systems, vibration signals and temperature field data exhibit complementary but asynchronous characteristics, while the actual fault diagnosis path requires feature fusion and joint analysis across data domains.

[0063] In the continuous decision-making process for multi-level risk warning, there is a need for intelligent decision-making, such as dynamic adjustment of feature weights and optimization of adjustment actions. If multi-feature weighted fusion and digital twin verification are not carried out, it is easy to cause insufficient protection or over-response.

[0064] In a multi-module collaborative diagnostic model, the vibration spectrum, time-domain statistical characteristics, and temperature field distribution within the same time window may be associated with different fault modes. It is necessary to fuse and understand the multi-dimensional feature nodes to accurately generate risk assessment indicators.

[0065] The system and method proposed in this application do not rely on single threshold alarms, fixed rule judgments or traditional signal processing methods. Instead, they take multi-sensor data fusion as the core and complete dynamic risk assessment and adaptive protection control through structural feature extraction, weighted combination and digital twin simulation mechanism.

[0066] It should be noted that the system described in this application is applicable to industrial equipment condition monitoring scenarios with one of the following technical complexity characteristics:

[0067] The current monitoring data does not explicitly characterize all fault features, and diagnostic conclusions need to be obtained through multi-feature fusion;

[0068] The sensor signal contains the evolution trend of the previous time series in a new round of acquisition, which cannot be inferred from the instantaneous value alone;

[0069] The fault development path exhibits multimodal characteristics that intersect, mutually corroborate, or contradictory;

[0070] It is understood that this application can also be extended to other industrial equipment condition monitoring scenarios with complex multi-physics field monitoring needs, including but not limited to health management systems for rotating machinery such as ship propulsion systems, large compressor units, and wind turbine generator sets.

[0071] Please see Figure 1 The present invention provides an embodiment of a vibration isolation status detection device for marine diesel generator sets, comprising the following steps: an acquisition module, a spectrum analysis module, a statistical analysis module, a temperature field module, a comprehensive module, and a twin response module;

[0072] The acquisition module uses a configured sensor array combined with a filter for signal preprocessing to acquire a first real-time parameter sequence, a second real-time parameter sequence, and a third real-time parameter sequence of the ship's diesel generator set. The first real-time parameter sequence characterizes the vibration state of the ship's generator set itself, the second real-time parameter sequence characterizes the vibration state of the ship's generator set foundation or vibration isolation system, and the third real-time parameter sequence characterizes the temperature field distribution state of the key area of ​​the vibration isolation system.

[0073] The spectrum analysis module calculates a first characteristic value characterizing the energy distribution of the vibration spectrum of the unit body based on the first real-time parameter sequence; the first characteristic value is used to indicate the risk of rotor imbalance or misalignment during unit operation.

[0074] The statistical analysis module calculates a second characteristic value representing the time-domain statistical characteristics of the foundation vibration based on the second real-time parameter sequence; the time-domain statistical characteristics are related to the random interference between the unit vibration and the external environment vibration, and are used to indicate the risk of vibration isolator impact or bolt loosening.

[0075] The temperature field module calculates a third characteristic value characterizing the uniformity of the temperature field distribution and the intensity of hot spots based on the third real-time parameter sequence; the third characteristic value is used to indicate the risk of abnormal temperature rise caused by at least one of the following abnormalities: overload abnormality, aging abnormality, or internal friction abnormality.

[0076] The comprehensive module generates a real-time risk index characterizing the current vibration isolation status and health risk level by weighting and combining the first, second, and third feature values. It should be further explained that the weighting in the comprehensive module of this embodiment is obtained through the following process: First, historical fault data and normal operation data are organized to construct a sample dataset containing different fault types (including rotor imbalance, misalignment, vibration isolator impact, bolt loosening, and abnormal temperature rise), fault severity, and corresponding first, second, and third feature values. The feature values ​​in the sample data are standardized to eliminate dimensional differences. Principal component analysis is used to mine the correlation between each feature value and the probability of fault occurrence and the degree of fault impact, calculating the initial objective weight of each feature value for risk assessment, and then combining it with the relevant data. Domain experts, recognizing the severity levels and susceptibility of different fault types and eigenvalues, use the analytic hierarchy process (AHP) to refine the initial objective weights multiple times, determining the basic weight coefficients. A dynamic adjustment model is constructed based on the unit's real-time operating load and speed. The basic weight coefficients are fine-tuned in real-time according to the impact of operating condition changes on the effectiveness of fault characterization of each eigenvalue, forming the final dynamic weight set. This dynamic weight set is used, on one hand, in the comprehensive module to perform a weighted summation of the first, second, and third eigenvalues, generating a real-time risk indicator characterizing the current vibration isolation status and health risk level. On the other hand, when determining an abnormal vibration isolation status, this embodiment combines the abnormality degree value of the vibration isolation status transformed from the real-time risk indicator with a preset eigenvalue-fault type-abnormality degree mapping table to obtain the abnormality degree corresponding to the current global abnormality type combination set, including:

[0077] Based on the real-time risk index value, it is converted into a vibration isolation state anomaly value in the range of 0 to 1 through a preset linear mapping function; the input of the linear mapping function is the real-time risk index value, and the output is the vibration isolation state anomaly value. Its conversion formula is determined based on the safe operation lower limit value and dangerous operation upper limit value of the real-time risk index.

[0078] Based on the vibration isolation state anomaly degree value, the first feature value, the second feature value, the third feature value, and the dynamic weight set used in the integrated module, the contribution ratio of each feature value in the global anomaly is calculated using the weighted Euclidean distance algorithm. Specifically, the deviation of each feature value from its healthy baseline value is first weighted according to the dynamic weight set, and then the ratio of the weighted deviation of each feature value to the sum of the weighted deviations of all feature values ​​is calculated to obtain the anomaly contribution ratio corresponding to each feature value.

[0079] Based on the calculated abnormal contribution ratio of each feature value, the preset feature value-fault type-abnormality mapping table is queried. The mapping table is stored in vector form, and each record contains a set of feature value contribution ratio distribution patterns, corresponding fault type combination labels, and pre-labeled comprehensive abnormality value. The K-nearest neighbor algorithm is used to match the current feature value contribution ratio vector with the historical vectors in the mapping table. The average value of the comprehensive abnormality value corresponding to the K historical records with the smallest Euclidean distance to the current vector is determined as the abnormality value corresponding to the current global abnormality type combination set.

[0080] The twin response module is used to respond to the real-time risk indicator meeting a preset trigger condition. It selects the adjustment action corresponding to the trigger condition from a set of actions including at least three adjustment actions, and combines this with a twin simulation algorithm to perform real-time simulation and real-time risk indicator monitoring and adjustment. The adjustment action includes at least issuing a warning signal, issuing an alarm signal, performing a load reduction operation, or performing a unit switching operation.

[0081] Further explanation is needed; please refer to [link / reference]. Figure 2 This embodiment uses a configured sensor array combined with a filter for signal preprocessing, including:

[0082] The original vibration signals and original infrared images of the marine diesel generator set are simultaneously acquired using a configured sensor array. The original vibration signals include signals from a first vibration sensor fixed to the generator set body and signals from a second vibration sensor fixed to the generator set foundation or vibration isolation system. It should be further noted that this embodiment simultaneously acquires the original infrared images of the marine diesel generator set and obtains the original temperature signals using an infrared thermal imager, specifically including:

[0083] Based on the spatial structure model of the key area of ​​the vibration isolation system, an infrared thermal imager is set up in the orthogonal lateral orientation of the key area; by adjusting the pitch and horizontal angles of the infrared thermal imager, its instantaneous field of view completely covers the three-dimensional monitoring area formed by the vibration isolator bearing interface, the body deformation area and the connecting fasteners.

[0084] Based on the same time-base synchronization signal as the vibration signal acquisition system, the infrared thermal imager is controlled to continuously acquire raw infrared radiation data of the target area at a preset frame rate; the raw infrared radiation data acquired in each frame is converted from analog to digital to generate a two-dimensional matrix composed of pixel grayscale values; multiple two-dimensional matrices that are continuously output in time sequence are combined to form a raw infrared image sequence.

[0085] Based on the original infrared image sequence, the radiation-temperature conversion parameters pre-calibrated by the infrared thermal imager are invoked; through the radiation thermometry algorithm built into the infrared thermal imager, the grayscale value of each pixel in each two-dimensional matrix in the original infrared image sequence is converted into the corresponding absolute temperature value in real time; thereby generating a sequence composed of two-dimensional matrices containing absolute temperature values, which is the original temperature signal characterizing the temperature field distribution state of the key area of ​​the vibration isolation system.

[0086] Based on the original vibration signal, impedance matching and signal amplification are performed using a charge amplifier, and synchronous sampling is performed using a 16-bit precision analog-to-digital converter at a sampling rate no less than a preset multiple of the highest analysis frequency to obtain a discrete digital vibration sequence. It should be further explained that the principle behind setting the preset sampling rate multiple to no less than 2.5 in this embodiment is based on the Nyquist sampling theorem, which states that for distortion-free signal reconstruction, the sampling frequency must be higher than twice the highest frequency of the signal. Considering the specific characteristics of vibration monitoring of marine diesel generator sets in this scenario, this setting aims to provide sufficient transition band for the anti-aliasing filter to suppress spectral aliasing, while retaining the analytical capability for high-frequency impact features such as bearing faults, and providing engineering safety margins for complex operating conditions. In specific implementation, the highest analysis frequency is dynamically calculated based on the unit's maximum continuous operating speed and a preset feature order multiple. Then, the highest analysis frequency is multiplied by a fixed preset multiple to adaptively configure the sampling frequency of the analog-to-digital converter, thereby ensuring that the vibration signal is acquired without aliasing across the entire operating range, providing a reliable data foundation for subsequent multi-dimensional feature extraction and fault diagnosis.

[0087] Based on the discretized digital vibration sequence, a segment of the signal under stationary conditions of the equipment is extracted or the stopband characteristics of a high-pass filter are used to calculate the DC component of the discretized digital vibration sequence.

[0088] The DC component of the discretized digital vibration sequence is subtracted from each data point in the discretized digital vibration sequence to obtain a vibration signal with zero bias. It should be further explained that the principle behind using both stationary signal extraction and high-pass filter stopband characteristics to calculate the DC component in this embodiment is that marine diesel generator sets are subject to environmental interference such as continuous low-frequency swaying of the hull during operation, causing baseline drift in the vibration signal and directly affecting the accuracy of time-domain statistical features. Specifically, during the stationary phase before generator start-up, the system automatically acquires a baseline signal and calculates its arithmetic mean as the reference DC component. During continuous operation, the real-time signal is initially filtered using a high-pass filter stopband, and the filtered low-frequency components are used as a reference value for the dynamic DC component. By subtracting this DC component from the original signal, it is ensured that subsequent vibration feature extraction is not affected by baseline drift.

[0089] The zero-bias vibration signal is input into a fourth-order Butterworth low-pass filter with a cutoff frequency set to a preset multiple of the unit's highest operating frequency. This filters out high-frequency vibration signals above the unit's highest operating frequency, resulting in a first pre-processed vibration signal. It should be further explained that in this embodiment, a fourth-order Butterworth low-pass filter with a cutoff frequency set to 2.5 times the unit's highest operating frequency is used. Its principle is based on the Nyquist sampling theorem to establish an anti-aliasing protection mechanism, retaining the effective fault characteristic frequency band below the highest operating frequency while filtering out high-frequency noise and interference components above the highest operating frequency. Specifically, the highest operating frequency is dynamically calculated based on the unit's rated speed, and the cutoff frequency is set to 2.5 times the highest operating frequency. Utilizing the characteristics of the fourth-order Butterworth filter—maintaining amplitude stability in the passband and providing sufficient attenuation in the stopband—the vibration signal is pre-processed through recursive calculations using the differential equations of a digital filter.

[0090] The first preprocessed vibration signal is input into a linear-phase FIR bandpass filter that includes the passband frequency range constructed from the fundamental frequency corresponding to the rated speed of the unit and a Hanning window function to obtain a first real-time parameter sequence and a second real-time parameter sequence. It should be further explained that in this embodiment, a Hanning window linear-phase FIR bandpass filter with a passband range of [0.7f0, ​​15f0] is constructed based on the fundamental frequency f0 of the rated speed of the unit. This embodiment retains the fundamental frequency harmonic band where the fault characteristic frequencies such as rotor imbalance and misalignment are located, while filtering out low-frequency interference such as hull sway and high-frequency components such as electromagnetic noise. In specific implementation, the fundamental frequency f0 is calculated according to the rated speed, and a 200th-order FIR filter is designed using a Hanning window to ensure linear phase characteristics to avoid waveform distortion. Vibration features directly related to mechanical faults are extracted through convolution operations, thereby obtaining pure first and second real-time parameter sequences.

[0091] Based on pre-calibrated reference data using a standard blackbody at different temperature points, a two-point correction calculation is performed on each frame of the original infrared image to obtain an initial calibrated infrared image. Specifically, this involves: selecting the original pixel response values ​​V1 and V2 corresponding to two standard temperature reference points T1 and T2; calculating the correction parameters for each pixel using the formula gain coefficient G=(T2-T1) / (V2-V1) and offset O=T1-G×V1; linearly transforming the original value V of each pixel using the calibration formula T_corrected=G×V+O; and outputting the non-uniformity-corrected temperature field image pixel by pixel; where T_corrected represents the calibrated pixel value.

[0092] Based on the initial calibrated infrared image, a two-dimensional convolution operation is performed using a 3×3 pixel Gaussian convolution kernel to obtain a spatially smoothed temperature field image. The weight coefficients of the Gaussian convolution kernel are calculated based on a two-dimensional Gaussian function with a standard deviation σ=0.8. It should be further noted that in this embodiment, a 3×3 Gaussian convolution kernel with a standard deviation σ=0.8 is used to spatially smooth the infrared image. The principle is to utilize the weight distribution characteristics of the Gaussian function to effectively suppress spatial high-frequency noise caused by water vapor interference and sensor noise while preserving the true temperature gradient characteristics. Specifically, by constructing a standardized weight coefficient matrix and performing a two-dimensional convolution operation with the calibrated temperature field image, a weighted average is achieved within a 5×5 pixel neighborhood. This eliminates isolated noise points without significantly blurring hotspot boundaries, providing temperature field data with optimized signal-to-noise ratio for subsequent time-domain analysis and hotspot detection.

[0093] The process of independently performing temporal median filtering on each pixel location in the spatially smoothed temperature field image yields a preprocessed third real-time parameter sequence. This process includes: obtaining the temperature value sequence of the current pixel over the past N sampling periods, forming a time window of length N; sorting all temperature values ​​within the time window and taking its median value; and finally replacing the original temperature value of the current pixel with this median value. It should be further noted that in this embodiment, the window length N is adaptively adjusted according to the real-time load change rate ΔP of the unit. Specifically, when |ΔP|>10% rated power / minute, N=5; when |ΔP|≤10% rated power / minute, N=11. This nonlinear filtering effectively suppresses transient temperature drift, ultimately outputting a stable third real-time parameter sequence.

[0094] Existing technologies that directly perform full-band energy statistics mask the proportion of characteristic frequency band energy related to rotor imbalance and misalignment, causing fault characteristics to be drowned out by background noise. Insufficient frequency resolution and window function mismatch in spectrum analysis cause spectral energy diffusion, distorting the true energy representation at characteristic frequencies. Failure to distinguish between steady-state energy and transient impact energy makes it difficult to capture transient spectral anomalies in the early stages of a fault. These problems collectively lead to decreased fault diagnosis accuracy, insufficient early warning capabilities, and reduced reliability of subsequent protection decisions. Therefore, please refer to [link to relevant documentation]. Figure 3 It should be further explained that, in this embodiment, the calculation of the first characteristic value characterizing the energy distribution of the body's vibration spectrum includes:

[0095] Based on the first real-time parameter sequence, the time-domain vibration signal is converted into a frequency-domain spectrum using the fast Fourier transform algorithm to obtain complex spectrum data containing amplitude and phase information;

[0096] Based on the complex spectrum data and a preset frequency band feature division rule, a set of frequency band intervals is obtained; the set of frequency band intervals includes the fundamental frequency band, the second octave band, and the high-frequency feature band; wherein the fundamental frequency band is the frequency at the rated rotational speed. Centered on [0.9], the frequency range is [0.9]. 1.1 The narrowband spectrum is used to capture rotor imbalance; the second harmonic is based on twice the fundamental frequency. Centered on [1.9], the frequency range is [1.9]. 2.1 The narrowband spectrum of ] is used to capture misalignment; the high-frequency characteristic band is the frequency range of [5]. 15 The broadband spectrum of [ ] is used to capture high-frequency shocks in bearings / gears;

[0097] Integrate the squared amplitude of each frequency point in each frequency band of the frequency band set to obtain a frequency band energy value sequence; the frequency band energy value sequence includes fundamental band energy value, second harmonic band energy value and high-frequency characteristic band energy value.

[0098] Based on the weighted algorithm combining the fundamental band energy value, second harmonic band energy value, and high frequency characteristic band energy value, a first characteristic value characterizing the energy distribution of the body vibration spectrum is obtained.

[0099] Understandably, the above calculation process only calculates the first feature value of a detection location. However, a single monitoring point cannot pinpoint the specific fault location. Therefore, this embodiment provides a step for calculating the location by combining the first feature value characterizing the energy distribution of the vibration spectrum of the main body with multiple first real-time parameter sequences obtained by multiple vibration sensors distributed at the diesel engine output end, generator input end, and the middle of the unit. Specifically, this includes:

[0100] Based on multiple real-time parameter sequences obtained by multiple vibration sensors distributed at the diesel engine output end, generator input end and the middle of the unit, the first characteristic value of each measuring point is calculated respectively.

[0101] When the first characteristic value of any measuring point exceeds the preset threshold, the first characteristic values ​​of all measuring points are compared, and the location of the measuring point with the largest first characteristic value is determined as the approximate area of ​​the fault source.

[0102] Based on the identified fault area, perform the following fault type identification process:

[0103] If the fundamental frequency band energy value in the first characteristic value of the fault area is greater than the sum of the second harmonic band energy value and the high frequency characteristic band energy value, it is determined that the fundamental frequency band energy is dominant. Then, the fundamental frequency phase of the vibration sensor signal at both ends of the unit is extracted, and the phase difference is calculated. If the phase difference is close to 0° or 180°, it is determined to be a rotor imbalance fault.

[0104] If the second harmonic energy value in the first characteristic value of the fault area is greater than the sum of the fundamental frequency energy value and the high frequency characteristic band energy value, then the second harmonic phase of the vibration sensor signal at both ends of the unit is extracted, the phase difference is calculated, and if the phase difference is close to 180°, it is determined to be a shaft misalignment fault.

[0105] If the high-frequency characteristic band energy value in the first characteristic value of the fault area is greater than the sum of the fundamental band energy value and the second harmonic band energy value, then envelope spectrum analysis is performed on the vibration signal of the fault area, the identified characteristic frequencies are matched with the standard bearing fault frequency library, and the specific bearing component that has failed is determined based on the matching results.

[0106] This process effectively combines fault feature separation and precise location by constructing a diagnostic architecture that integrates multi-band energy analysis and multi-sensor positioning. In particular, by dividing the characteristic frequency bands corresponding to specific fault mechanisms and performing independent energy calculations, it overcomes the defect of feature information being submerged by background noise in full-band analysis, enabling the spectral characteristics of various faults to be clearly presented. Secondly, by adopting a strategy that combines multi-measurement point feature value comparison with phase analysis, it not only achieves rapid preliminary location of the fault area, but also achieves precise judgment of fault type and identification of specific components through phase relationship and envelope spectrum characteristics. This hierarchical and progressive diagnostic mechanism significantly improves the sensitivity of early fault detection and the reliability of diagnostic conclusions, providing solid technical support for subsequent intelligent maintenance decisions, and ultimately comprehensively enhancing the overall performance of the ship generator set condition monitoring system.

[0107] It should be further explained that, in this embodiment, the calculation of the second eigenvalue characterizing the time-domain statistical properties of foundation vibration includes:

[0108] Based on the second real-time parameter sequence and the configured fixed-duration sliding time window, multiple overlapping sub-time period signal sets are obtained; wherein the length of the fixed-duration sliding time window is set to 0.5 seconds and the overlap rate is set to 50%.

[0109] Based on the signal set of each sub-period, the time-domain statistical parameter set of each sub-period signal is calculated in parallel. The time-domain statistical parameter set includes the peak-to-peak value, kurtosis value, peak factor, and impulse index of the sub-period signal. It should be further noted that the selection of the four time-domain statistical parameters of peak-to-peak value, kurtosis value, peak factor, and impulse index in this embodiment is a comprehensive characterization scheme designed based on the physical characteristics of the vibration isolation system fault of the marine diesel generator set. Among them, the peak-to-peak value directly characterizes the overall fluctuation amplitude of the vibration signal by calculating the difference between the maximum and minimum values ​​of the signal within the sub-period, and is used to detect severe vibrations caused by vibration isolator impact or bolt loosening. The kurtosis value is calculated by the fourth-order central moment of the signal and... The ratio of the fourth power of the standard deviation is highly sensitive to impact signals and can effectively capture the non-Gaussian impact characteristics caused by faults such as bolt loosening; the peak factor, as the ratio of the peak value to the root mean square value, is used to distinguish between steady-state vibration and impact vibration, and this value increases significantly when the vibration isolator experiences intermittent impacts; the pulse index, through the ratio of the peak value to the absolute average value, further enhances the ability to identify periodic impacts; the combination of these four parameters comprehensively characterizes the time-domain statistical characteristics of foundation vibration from four dimensions: vibration intensity, impact characteristics, waveform morphology, and pulse periodicity, overcoming the shortcomings of incomplete characterization by a single parameter, and providing multi-angle quantitative basis for vibration isolator impact and bolt loosening risk;

[0110] Based on a preset interference component threshold rule, the time-domain statistical parameter set is filtered to obtain a filtered effective time-domain statistical parameter set; wherein the filtering process includes:

[0111] When the peak factor of the signal in a sub-period is less than 2.0, it is determined that the signal is dominated by low-frequency environmental interference, and all statistical parameters of that sub-period are removed.

[0112] When the peak-to-peak value of the signal in a sub-period is lower than the background noise threshold, it is determined to be an invalid vibration signal, and all statistical parameters of that sub-period are removed.

[0113] Based on the filtered effective time-domain statistical parameter set combined with the preset weighted contribution weight, the feature value of each effective sub-period is obtained. It should be further explained that the process of obtaining the weighted contribution weight in this embodiment includes: establishing a training sample set based on the vibration isolator impact and bolt loosening cases concentrated in the historical fault dataset; using principal component analysis to perform sensitivity analysis on each time-domain statistical parameter and calculate the contribution of each parameter to fault classification; then constructing an expert scoring system and combining the actual operating conditions of the ship diesel generator set to make multiple rounds of corrections to the initial contribution; finally, through verification tests under typical fault scenarios, the weight allocation scheme is iteratively optimized to ensure that the false alarm rate is effectively controlled while ensuring the fault detection rate, thereby determining the final weighted contribution weight allocation scheme. The effective sub-time period refers to a set of multiple overlapping sub-time period signals obtained based on a second real-time parameter sequence combined with a configured fixed-duration sliding time window. For each sub-time period signal set, a time-domain statistical parameter set including peak-to-peak value, kurtosis value, peak factor, and impulse index is calculated in parallel. Then, based on a preset interference component threshold rule, the time-domain statistical parameter set is filtered. Sub-time periods that simultaneously meet the validity conditions of having a peak factor of not less than 2.0 and a peak-to-peak value of not less than the background noise threshold, and are not judged and eliminated by the filtering rule, are considered valid. Specifically, the filtering rule is as follows: when the peak factor of a sub-time period signal is less than 2.0, it is determined to be dominated by low-frequency environmental interference, and all statistical parameters of that sub-time period are eliminated; when the peak-to-peak value of a sub-time period signal is less than the background noise threshold, it is determined to be an invalid vibration signal, and all statistical parameters of that sub-time period are eliminated. The time-domain statistical parameter set corresponding to the effective sub-time period is retained and used for subsequent calculation of the corresponding sub-time period's feature value based on a preset weighted contribution weight.

[0114] Based on the characteristic values ​​of all valid sub-time periods, the 95th percentile is taken as the second characteristic value characterizing the time-domain statistical characteristics of the foundation vibration. This embodiment selects the 95th percentile as the characterization value for the second characteristic value, a technical optimization based on the statistical characteristics of impact faults in marine diesel generator set vibration isolation systems. This quantile can effectively capture statistically significant high-intensity impact events in the vibration signal, while eliminating random single-point extreme value interference. Compared to the shortcomings of the maximum value being easily affected by noise or the average value being easily diluted by a large amount of stable data, the 95th percentile can reflect the typical intensity level of the fault impact while maintaining sensitivity to abnormal states, ensuring that the second characteristic value achieves the best balance between fault early warning reliability and anti-interference robustness.

[0115] This process achieves accurate monitoring and early warning of impact faults in vibration isolation systems through multi-dimensional time-domain parameter fusion and statistical optimization. In particular, by combining four parameters—peak-to-peak value, kurtosis, peak factor, and pulse index—a complete time-domain characterization system is constructed from four dimensions: vibration intensity, impact characteristics, waveform morphology, and pulse periodicity, overcoming the incompleteness of single-parameter characterization. Secondly, a weight optimization method based on principal component analysis and an expert scoring system is employed to ensure the scientific contribution of each parameter to fault diagnosis. Finally, 95th percentile statistical processing effectively suppresses accidental noise interference while preserving typical impact characteristics. This multi-level optimization design enables the system to maintain sensitivity to early faults while possessing good anti-interference capabilities, ultimately achieving highly reliable early warning of vibration isolator impact and bolt loosening faults.

[0116] It should be further explained that the third characteristic value, which characterizes the uniformity of the temperature field distribution and the intensity of hot spots, is calculated in this embodiment, including:

[0117] Based on the third real-time parameter sequence, an adaptive threshold segmentation algorithm is used to extract hotspot candidate regions in the temperature field image. The adaptive threshold is determined by calculating the inter-peak threshold of the image grayscale histogram. Further, this embodiment calculates the grayscale histogram of the spatially smoothed temperature field image and uses the maximum inter-class variance method to determine the optimal segmentation threshold for distinguishing hotspots from the background in the grayscale histogram. All pixels in the image are binarized according to the segmentation threshold to generate a binary image containing potential hotspot regions. Morphological closing operations are performed on the hotspot candidate regions. A 3×3 square structuring element is first dilated to fill holes, and then eroded to smooth the boundaries, obtaining a connected and complete set of effective hotspot regions.

[0118] Based on the set of effective hotspot regions, a set of hotspot feature parameters is calculated. These parameters include the percentage of the total area of ​​the hotspot regions and the difference between the highest temperature in a hotspot region and the average temperature across the entire region. Specifically, the process is as follows: First, the number of pixels in all effective hotspot regions is counted and multiplied by the actual area corresponding to each pixel. This sum is then accumulated to obtain the total area of ​​the effective hotspot regions. This total area is then divided by the total area of ​​the temperature field image to obtain the percentage of the total area of ​​the hotspot regions. Simultaneously, the maximum pixel temperature within each effective hotspot region is extracted as the highest temperature in the hotspot region. The arithmetic mean of the temperatures of all pixels in the temperature field image is calculated as the average temperature across the entire region. The highest temperature in the hotspot region is subtracted from the average temperature across the entire region to obtain the difference between the highest temperature in the hotspot region and the average temperature across the entire region. This completes the calculation of the hotspot feature parameters.

[0119] Based on the pixel temperature values ​​of the temperature field image, the coefficient of variation of the global temperature distribution is calculated to characterize the uniformity. The specific process includes: first, traversing the temperature values ​​of all pixels in the temperature field image, accumulating all pixel temperature values ​​and dividing by the total number of pixels to obtain the arithmetic mean of the global temperature; then, calculating the sum of squares of the differences between each pixel temperature value and the arithmetic mean, and dividing by the total number of pixels to obtain the variance of the global temperature; taking the square root of the variance to obtain the standard deviation of the global temperature; finally, dividing the standard deviation of the global temperature by the arithmetic mean to obtain the coefficient of variation of the global temperature distribution, thereby characterizing the uniformity of the temperature field distribution.

[0120] By combining preset uniformity weights and hotspot intensity weights, the coefficient of variation and the set of hotspot characteristic parameters are weighted and fused to obtain a third characteristic value characterizing the uniformity of the temperature field distribution and the intensity of the hotspots. It should be further noted that the process of obtaining the uniformity weights and hotspot intensity weights in this embodiment includes:

[0121] Based on the temperature field characteristics analysis of vibration isolator overload, aging, and internal friction anomaly cases in historical fault datasets, principal component analysis (PCA) was used to calculate the contributions of temperature field distribution uniformity and hot spot intensity to fault classification. The initial contributions were then refined multiple times using an expert scoring system. The weight allocation scheme was iteratively optimized through verification tests under typical fault scenarios, ultimately determining the specific weighting ratios for uniformity and hot spot intensity. This process, through multi-dimensional temperature field analysis and intelligent weight allocation, achieves accurate diagnosis and early warning of vibration isolator thermal faults. In particular, adaptive threshold segmentation and morphological processing enable precise extraction of hot spot regions, effectively overcoming the insensitivity of traditional temperature monitoring methods to local anomalies. Secondly, an evaluation system combining the coefficient of variation and hot spot characteristic parameters was adopted to construct complete evaluation indicators from two dimensions: temperature field distribution uniformity and local overheating intensity. Finally, the weight allocation scheme optimized by PCA and the expert system ensures the reasonable contribution of each evaluation indicator under different fault modes. This multi-level, quantitative temperature field analysis method significantly improves the accuracy of identifying faults such as overload, aging, and internal friction in vibration isolators, providing reliable technical support for the thermal safety protection of ship generator sets.

[0122] It should be further explained that this embodiment combines a twin simulation algorithm for real-time simulation and real-time risk indicator monitoring and adjustment, including:

[0123] The real-time risk index value, which represents the current vibration isolation status and health risk level, is compared with the preset risk index value threshold to make an overall anomaly judgment. When the real-time risk index value is greater than the preset risk index value threshold, the vibration isolation status is determined to be globally abnormal, an alarm is triggered, and the degree of vibration isolation status abnormality value is obtained based on the real-time risk index value.

[0124] When an anomaly is determined, the anomaly degree corresponding to the current global anomaly type combination set is obtained based on the vibration isolation state anomaly degree value, the first feature value, the second feature value, the third feature value, and a preset feature value-fault type-anomaly degree mapping table. It should be further noted that this embodiment analyzes the weights between the vibration isolation state anomaly degree value and the first, second, and third feature values ​​and the real-time risk indicator value to obtain the severity weights of the corresponding anomaly types under the first, second, and third feature values. The feature value-fault type-anomaly degree mapping table in this embodiment was obtained by those skilled in the art based on historical first, second, and third feature values, fault types, and anomaly... The feature value-fault type-anomaly degree mapping table is constructed by combining the degree of anomaly with a tree-structured database. The motivation for constructing the feature value-fault type-anomaly degree mapping table in this embodiment is to solve the problem of fault type identification and severity quantification in multi-feature value fusion diagnosis. Its core function is to transform discrete feature value indicators into a unified anomaly degree assessment by establishing a mapping relationship from a multi-dimensional feature space to a fault decision space, thereby overcoming the limitations of single threshold judgment in complex fault scenarios. This mapping table is constructed based on historical fault data and can accurately reflect the fault modes and their severity corresponding to different feature value combinations. It provides accurate type identification and level classification basis for subsequent protection decisions, and significantly improves the accuracy and reliability of fault diagnosis under complex working conditions.

[0125] Based on the anomaly severity corresponding to the current global anomaly type combination set, and combined with the anomaly type-adjustment action type-adjustment severity mapping table, a global candidate adjustment action parameter set is obtained. The adjustment severity includes the adjustment priority of each adjustment action, the adjustment time length of the corresponding adjustment action, and the size of the adjusted unit range. This process addresses the problems of adjustment actions corresponding to a single anomaly type being difficult to adapt to scenarios with multiple anomaly type combinations, and the lack of unified quantitative basis for the priority, time length, and unit range of adjustment actions, thus avoiding over- or under-intervention due to blind adjustment strategies. Its function is to associate the global anomaly type combination set with the anomaly severity, combined with preset anomaly types... The system uses an adjustment action type-adjustment degree mapping table to systematically organize the appropriate adjustment parameters corresponding to different anomaly combinations. It clarifies the execution priority, duration, and coverage range of adjustment actions, ensuring precise matching between adjustment actions and the overall anomaly state. This improves the targeting and orderliness of adjustment actions while reducing the impact of the adjustment process on the continuous operation of the ship's diesel generator sets. Based on historical handling experience and expert knowledge, this mapping table can automatically generate an optimized parameter set containing action priority, duration, and impact range according to the anomaly type. This provides a precise initial decision-making scheme for subsequent digital twin simulation, achieving intelligent closed-loop control from state diagnosis to protection execution.

[0126] Based on the global candidate adjustment action parameter set combined with the twin simulation model constructed from the marine diesel generator set, after abnormal simulation adjustment, a global inference response dataset is obtained. The specific steps for abnormal simulation adjustment based on the twin simulation model in this embodiment include: firstly, injecting the global candidate adjustment action parameter set into a multi-physics coupled digital twin containing a generator dynamics model, a vibration isolation system transmission model, and a thermodynamic model; then, setting initial boundary conditions consistent with real-time operating conditions, advancing the simulation process with a fixed time step, and dynamically solving the transient evolution data of generator vibration response, vibration isolation force transmissibility, and temperature field distribution after the adjustment action is executed; finally, outputting a global inference response dataset covering vibration spectrum, time-domain statistical characteristics, and temperature field gradient through a multi-physics data fusion engine.

[0127] Based on the global simulation response dataset and combined with a multi-objective optimization decision-making algorithm, the optimal set of adjustment actions is obtained with the goal of maximizing the preset comprehensive decision factor. In this embodiment, the comprehensive decision factor is obtained by weighting the comprehensive impact score and the comprehensive performance score. The comprehensive performance score is obtained by weighting the vibration suppression rate, temperature drop amplitude, and state stability recovery time measured in real-time. In this embodiment, the comprehensive impact score is obtained by weighting the execution time, equipment wear, and action execution risk probability of the optimal set of adjustment actions. In this embodiment, the vibration suppression rate is calculated based on the time-domain and frequency-domain response data output by the digital twin model, obtained by comparing the relative changes of the first and second characteristic values ​​before and after the adjustment action is executed. The temperature drop amplitude is obtained based on thermodynamic simulation, calculated by establishing a three-dimensional transient heat conduction model of the vibration isolation system to determine the temperature field of the hot spot area after the adjustment action is implemented. Gradient reduction; State stability recovery time is obtained by analyzing the system dynamic response curve in the digital twin simulation data, defined as the time from the start of the adjustment action to the system's key parameters entering the stable operating threshold range; Parameterization of execution time is based on the instruction response timing analysis of the equipment control system, obtained by decomposing the sub-steps in the action execution process and accumulating the standard operation time; Quantification of equipment loss adopts a cumulative damage algorithm based on the stress-life model to calculate the impact of mechanical and thermal stresses caused by the adjustment action on the fatigue life of the equipment, where the stress-life model is obtained by fitting a support vector machine; The probability of action execution risk is constructed through fault tree analysis, combined with historical operation and maintenance data to probabilistically model risk events such as equipment interlocking failures that may be caused during the action execution process. These parameters together constitute a comprehensive index system for evaluating adjustment actions from three dimensions: efficiency, cost, and risk.

[0128] In this embodiment, the process of obtaining the globally optimal set of adjustment action combinations based on a multi-objective optimization decision algorithm includes: taking the simulation and monitoring data corresponding to various adjustment action combinations contained in the global simulation response dataset as algorithm input; firstly, based on the time-domain and frequency-domain response data output by the digital twin model, calculating the vibration suppression rate by comparing the relative changes of the first and second characteristic values ​​before and after the adjustment action execution; calculating the temperature drop amplitude by relying on thermodynamic simulation and the three-dimensional transient heat conduction model of the vibration isolation system; analyzing the system dynamic response curve in the digital twin simulation data and statistically analyzing the time from the start of the adjustment action to the system's key parameters entering the stable operating threshold range to obtain the state stability recovery time; and then generating a comprehensive performance score through weighted processing of the three factors; and finally, controlling the equipment... The system analyzes the timing of command response, decomposes the action execution sub-steps, and accumulates the standard operation time to obtain the execution time. It uses a support vector machine-fitted stress-life model and cumulative damage algorithm to calculate the impact of mechanical and thermal stresses caused by adjustment actions on the fatigue life of equipment to quantify equipment wear. It uses fault tree analysis combined with historical operation and maintenance data to probabilistically model risk events such as equipment interlocking failures that may be caused by action execution to obtain the probability of action execution risk. The three factors are weighted to generate a comprehensive impact score. Finally, the comprehensive efficiency score and the comprehensive impact score are weighted and fused according to preset weights to obtain a comprehensive decision factor. The multi-objective optimization decision algorithm uses iterative calculation to select the optimal combination of adjustment actions in terms of efficiency, cost, and risk three-dimensional evaluation indicators, and finally obtains the globally optimal set of adjustment action combinations.

[0129] When the real-time risk indicator value is less than or equal to the preset risk indicator value threshold and at least one of the first feature value, the second feature value, and the third feature value is greater than the preset abnormality judgment threshold, the vibration isolation state is judged to be locally abnormal and an alarm is triggered.

[0130] Based on the feature value set corresponding to the local anomaly in the vibration isolation state, the feature value magnitude is combined with the feature value-fault type-anomaly degree mapping table to obtain the anomaly degree corresponding to the current local anomaly type combination set.

[0131] Based on the anomaly degree corresponding to the current set of local anomaly type combinations, combined with the anomaly type-adjustment action type-adjustment degree mapping table, a set of local candidate adjustment action parameters is obtained;

[0132] Based on the local candidate adjustment action parameter set, the local optimal adjustment action combination set is obtained by repeating the twin simulation and multi-objective optimization process.

[0133] When the real-time risk indicator value is less than or equal to the preset risk indicator value threshold, and the first, second, and third characteristic values ​​are all less than the preset anomaly judgment characteristics, the changing trends of the first, second, and third characteristic values ​​are monitored in real time according to the preset data interval. When the difference between at least one characteristic value and the corresponding preset anomaly judgment threshold is less than the preset risk warning threshold, an early warning operation is performed. At the same time, the warning level is determined according to the rate of change of the corresponding characteristic value, and a real-time warning of the corresponding level is issued. The motivation for setting up this early warning mechanism in this application is to realize the transformation of equipment status monitoring from post-event alarm to pre-event warning mode. By tracking the degree of approach and changing trend of characteristic values ​​relative to the anomaly threshold in real time, an early warning signal is issued in advance when obvious signs of deterioration have appeared before the fault has fully formed, thereby breaking through the lag limitation of traditional fixed threshold detection methods and gaining valuable response time windows for preventive maintenance decisions. The setting of the preset data interval in this application is comprehensively determined based on the physical time constant of the changes in the operating status of the ship's diesel generator set and the requirements for the timeliness of the early warning. The specific technical means are as follows: First, based on historical data statistical analysis, the typical time span required for the unit to deteriorate from normal state to the preset risk warning threshold is taken as one-tenth as the benchmark interval; second, the benchmark interval is dynamically adjusted in combination with the unit's current real-time load rate, automatically shortening the interval under high load conditions to increase monitoring density, and appropriately extending the interval under low load steady-state conditions to optimize system resources; finally, the preset data interval is set as a dynamic value that is adaptively associated with the unit's operating status, thereby ensuring that the trend of state deterioration can be captured in a timely manner, while avoiding unnecessary calculation and communication loads.

[0134] It should be further explained that the set of adjustment actions and the specific steps of each action in this embodiment include: issuing a warning signal includes real-time tracking of the difference between each feature value and the corresponding anomaly judgment threshold. When the difference is less than the preset risk warning threshold, the rate of change of the feature value is continuously monitored, and the warning level is determined according to the speed of change. For low-level warnings, a low-frequency audible and visual alarm is activated, and a text prompt is displayed in a prominent position on the monitoring platform, clearly indicating the risk type that triggered the warning and the change of the corresponding feature value. For medium-level warnings, a medium-frequency audible and visual alarm is activated, and a prominent window prompt pops up on the monitoring platform. At the same time, a message containing details of abnormal feature values, risk contribution ratio, and potential impact is pushed to the mobile application of the operation and maintenance personnel to ensure that the operation and maintenance personnel can quickly grasp the core information. Alarm signals are issued based on the severity of a global anomaly and the corresponding alarm level. A Level 1 alarm triggers a high-frequency audible and visual alarm, and the monitoring platform displays a red warning pop-up window, showing the anomaly type, involved characteristic values, and current risk level in detail. At the same time, a text message notification containing key anomaly information is sent to the maintenance personnel. A Level 2 alarm triggers a continuous audible and visual alarm, quickly notifying the maintenance personnel via text message, and then notifying the maintenance manager by telephone, clearly conveying the urgency of the anomaly, its scope of impact, and preliminary handling suggestions. After a local anomaly is determined, an alarm matching its severity is triggered, highlighting the set of characteristic values ​​that caused the local anomaly and the corresponding fault type in the pop-up window and notification information. The operation to reduce unit load includes two categories: basic load reduction and deep load reduction. Before basic load reduction, the risk type corresponding to the anomaly is identified, and an appropriate load reduction rate plan is formulated accordingly to smoothly and gradually reduce the unit load and avoid the impact of sudden load changes on the equipment. Deep load reduction further reduces the load scale on the basis of basic load reduction and shuts down non-core auxiliary loads in a preset priority order to reduce energy consumption and equipment operating pressure. During the load reduction process, relevant parameters such as vibration and temperature are continuously collected, and the changing trend of each characteristic value is monitored in real time. If the characteristic value does not show signs of improvement or continues to deteriorate, the action upgrade process is triggered to switch to a higher level of load reduction operation or other appropriate actions. The unit switching operation includes two types: standard switching and forced switching. Standard switching strictly follows a preset procedure. First, the standby unit is started, and the operating parameters of the standby unit, such as speed and voltage, are collected in real time by sensors and accurately compared with the operating parameters of the original unit. After the parameters are fully synchronized, the load transfer ratio is smoothly adjusted by the control system to ensure that the load gradually transitions from the original unit to the standby unit. After the load transfer is completed, the power supply circuit of the original unit is cut off, and the original unit is locked to prevent unauthorized start commands. Forced switching aims to quickly curb the deterioration of risks. It simplifies some unnecessary transition procedures, accelerates the start-up and parameter synchronization process of the standby unit, completes the load transfer in the shortest possible time, and immediately starts the cooling system of the original unit to prevent the vibration isolators from degrading or failing due to high temperature through continuous heat dissipation.During all adjustment actions, the sensor array continuously collects real-time data on unit vibration, foundation and vibration isolation system vibration, and temperature field distribution in key areas of the vibration isolation system. After preprocessing such as charge amplifier amplification, filter filtering, and analog-to-digital conversion, the first, second, and third characteristic values ​​are recalculated. The comprehensive module then weights and combines these values ​​again to generate real-time risk indicators. If the indicators drop to the target safe range, the current action status is maintained for a period of time, during which the system stability recovery time is continuously monitored through the twin model. Once the target is met, the normal monitoring mode is gradually restored. If the indicators do not reach the target range or rebound, the adjustment actions are upgraded according to the level gradient based on the iterative results of the multi-objective optimization decision algorithm. If equipment failure or abnormal command response is detected during the action execution, the emergency plan is immediately triggered to quickly reduce the unit load to a safe level. At the same time, maintenance personnel are urgently notified through multiple channels, providing detailed information on the fault tree analysis-based fault type, equipment status combined with real-time parameters, and emergency response suggestions. This forms a complete closed-loop mechanism of characteristic value monitoring, anomaly judgment, mapping table matching, twin simulation optimization, action execution, status feedback, and adjustment iteration, ensuring that risks are accurately and efficiently controlled.

[0135] This embodiment effectively overcomes the limitations of single threshold judgment in complex fault scenarios by constructing a global / local anomaly classification and judgment mechanism, combined with a feature value-fault type-anomaly degree mapping table and anomaly type-adjustment action type-adjustment degree mapping table built based on historical data and expert knowledge. It achieves accurate identification of fault types and quantitative assessment of anomaly degree under multi-feature value fusion, while also solving the problems of poor adaptability of adjustment actions and lack of unified quantitative basis for parameters in multi-anomaly combination scenarios. Relying on a multi-physics coupled digital twin model for anomaly simulation adjustment and global inference response data acquisition, and using a multi-objective optimization decision algorithm to comprehensively evaluate the effectiveness, cost, and risk of adjustment actions from three dimensions, it ensures globally / locally optimal adjustment actions. The scientific selection of the set avoids excessive or insufficient intervention caused by the blindness of the adjustment strategy, and significantly improves the pertinence, orderliness and safety of the adjustment actions. By setting up a pre-warning mechanism based on the degree of approximation and rate of change of characteristic values, the lag limitation of traditional fixed threshold detection is broken, realizing the transformation of equipment status monitoring from post-warning alarm to pre-warning mode, and gaining sufficient response time for preventive maintenance. The above technical means form an intelligent closed-loop control from anomaly judgment, fault identification, action matching, simulation optimization to early warning protection, which ultimately greatly improves the accuracy and reliability of fault diagnosis of marine diesel generator set vibration isolation system under complex working conditions, ensures the continuity, stability and safety of unit operation, and reduces equipment wear and maintenance costs.

[0136] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the present invention. All of these variations are within the protection scope of the present invention.

Claims

1. A vibration isolation condition testing device for marine diesel generator sets, characterized in that, include: Acquisition module, spectrum analysis module, statistical analysis module, temperature field module, comprehensive module, and twin response module; The acquisition module uses a configured sensor array combined with a filter to perform signal preprocessing to obtain the first real-time parameter sequence, the second real-time parameter sequence, and the third real-time parameter sequence of the ship's diesel generator set. The spectrum analysis module calculates a first characteristic value characterizing the energy distribution of the body vibration spectrum based on the first real-time parameter sequence. The statistical analysis module calculates a second characteristic value representing the time-domain statistical characteristics of the foundation vibration based on the second real-time parameter sequence. The temperature field module calculates a third characteristic value characterizing the uniformity of the temperature field distribution and the intensity of hot spots based on the third real-time parameter sequence. The integrated module generates a real-time risk index characterizing the current vibration isolation status and health risk level by weighting and combining the first feature value, the second feature value, and the third feature value. The twin response module is used to respond to the real-time risk indicator meeting the preset trigger condition by selecting the adjustment action corresponding to the trigger condition from the action set including at least three adjustment actions and combining it with the twin simulation algorithm to perform real-time simulation and real-time risk indicator monitoring and adjustment.

2. The vibration isolation condition detection equipment for marine diesel generator sets as described in claim 1, characterized in that, The first real-time parameter sequence characterizes the vibration state of the ship's generator set, the second real-time parameter sequence characterizes the vibration state of the generator set's foundation or vibration isolation system, and the third real-time parameter sequence characterizes the temperature field distribution in key areas of the vibration isolation system. The first characteristic value is used to indicate the risk of rotor imbalance or misalignment during generator set operation. The time-domain statistical characteristics are related to the random interference between the generator set vibration and the external environment vibration, and are used to indicate the risk of vibration isolator impact or bolt loosening. The third characteristic value is used to indicate the risk of abnormal temperature rise caused by at least one of the following abnormalities: abnormal overload, abnormal aging, or abnormal internal friction. The adjustment actions include at least issuing a warning signal, issuing an alarm signal, performing a generator set load reduction operation, or performing a generator set switching operation.

3. The vibration isolation condition detection equipment for marine diesel generator sets as described in claim 2, characterized in that, Signal preprocessing is performed using a configured sensor array combined with filters, including: The original vibration signals and original infrared images of the ship's diesel generator set are simultaneously acquired through a configured sensor array. Based on the original vibration signal, impedance matching and signal amplification are performed through a charge amplifier, and synchronous sampling is performed using a 16-bit precision analog-to-digital converter at a sampling rate no less than a preset multiple of the highest analysis frequency to obtain a discrete digital vibration sequence. Based on the discretized digital vibration sequence, a segment of the signal under stationary conditions of the equipment is extracted or the stopband characteristics of a high-pass filter are used to calculate the DC component of the discretized digital vibration sequence. The DC component of the discretized digital vibration sequence is subtracted from each data point in the discretized digital vibration sequence to obtain a vibration signal with zero bias.

4. The vibration isolation condition detection equipment for marine diesel generator sets as described in claim 3, characterized in that, Signal preprocessing using a configured sensor array combined with filters also includes: The zero-bias vibration signal is input into a fourth-order Butterworth low-pass filter with a cutoff frequency set to a preset multiple of the unit's highest operating frequency to filter out high-frequency vibration signals above the unit's highest operating frequency, thereby obtaining the first pre-processed vibration signal. The first preprocessed vibration signal is input into a linear phase FIR bandpass filter that includes the passband frequency range constructed from the fundamental frequency corresponding to the rated speed of the unit and the Hanning window function to obtain the first real-time parameter sequence and the second real-time parameter sequence. Based on the reference data calibrated in advance at different temperature points using a standard reference blackbody, a two-point correction calculation is performed on each frame of the input raw infrared image to obtain the initial calibrated infrared image; Based on the initial calibrated infrared image, a two-dimensional convolution operation is performed using a 3×3 pixel Gaussian convolution kernel to obtain a spatially smooth temperature field image. The temporal median filter is applied independently to each pixel location in the spatially smoothed temperature field image to obtain the preprocessed third real-time parameter sequence.

5. The vibration isolation condition detection equipment for marine diesel generator sets as described in claim 4, characterized in that, Calculate the first eigenvalue characterizing the energy distribution of the body's vibration spectrum, including: Based on the first real-time parameter sequence, the time-domain vibration signal is converted into a frequency-domain spectrum using the fast Fourier transform algorithm to obtain complex spectrum data containing amplitude and phase information; Based on the complex spectrum data and the preset frequency band feature division rules, a set of frequency band intervals is obtained; the set of frequency band intervals includes the fundamental frequency band, the second octave band, and the high-frequency feature band. Integrate the squared amplitude of each frequency point in each frequency band of the frequency band set to obtain a frequency band energy value sequence; the frequency band energy value sequence includes fundamental band energy value, second harmonic band energy value and high-frequency characteristic band energy value. Based on the weighted algorithm combining the fundamental band energy value, the second harmonic band energy value, and the high-frequency characteristic band energy value, a first characteristic value characterizing the energy distribution of the body's vibration spectrum is obtained.

6. The vibration isolation condition detection equipment for marine diesel generator sets as described in claim 5, characterized in that, The calculation of the second eigenvalue, which characterizes the time-domain statistical properties of foundation vibration, includes: Based on the second real-time parameter sequence combined with the configured fixed-duration sliding time window, multiple overlapping sub-time period signal sets are obtained; Based on the signal set of each sub-time period, the time-domain statistical parameter set of each sub-time period signal is calculated in parallel: the time-domain statistical parameter set includes the peak-to-peak value, kurtosis value, peak factor and impulse index of the sub-time period signal; Based on the preset interference component threshold rules, the time-domain statistical parameter set is filtered to obtain the filtered effective time-domain statistical parameter set. Based on the filtered effective time-domain statistical parameter set combined with the preset weighted contribution weight, the feature value of each effective sub-time period is obtained. Based on the characteristic values ​​of all valid sub-time periods, the 95th percentile is taken as the second characteristic value characterizing the time-domain statistical properties of the foundation vibration.

7. The vibration isolation condition detection equipment for marine diesel generator sets as described in claim 6, characterized in that, The calculation of the third eigenvalue, which characterizes the uniformity of the temperature field distribution and the intensity of hot spots, includes: Based on the third real-time parameter sequence, an adaptive threshold segmentation algorithm is used to extract hotspot candidate regions in the temperature field image; the adaptive threshold is determined by calculating the inter-peak threshold of the image grayscale histogram. Perform morphological closing operations on the hotspot candidate regions to obtain a set of effective hotspot regions; Based on the set of effective hotspot regions, a set of hotspot feature parameters is calculated; the set of hotspot feature parameters includes the percentage of the total area of ​​hotspot regions and the difference between the highest temperature of the hotspot region and the average temperature of the entire region. Based on the pixel temperature values ​​of the temperature field image, the coefficient of variation of the global temperature distribution is calculated to characterize the uniformity. By combining preset uniformity weights and hotspot intensity weights, the coefficient of variation and hotspot characteristic parameter set are weighted and fused to obtain a third characteristic value that characterizes the uniformity of the temperature field distribution and the intensity of the hotspots.

8. The vibration isolation condition detection equipment for marine diesel generator sets as described in claim 7, characterized in that, Combining twin simulation algorithms for real-time simulation and real-time risk indicator monitoring and adjustment, including: The real-time risk index value, which represents the current vibration isolation status and health risk level, is compared with the preset risk index value threshold to make an overall anomaly judgment. When the real-time risk index value is greater than the preset risk index value threshold, the vibration isolation status is determined to be globally abnormal, an alarm is triggered, and the degree of vibration isolation status abnormality value is obtained based on the real-time risk index value. When an anomaly is determined, the anomaly degree corresponding to the current global anomaly type combination set is obtained by combining the vibration isolation state anomaly degree value, the first feature value, the second feature value, the third feature value, and the preset feature value-fault type-anomaly degree mapping table. Based on the anomaly severity corresponding to the current global anomaly type combination set, combined with the anomaly type-adjustment action type-adjustment severity mapping table, a global candidate adjustment action parameter set is obtained; the adjustment severity includes the adjustment priority of each adjustment action, the adjustment time length of the corresponding adjustment action, and the size of the adjustment unit range.

9. The vibration isolation condition detection equipment for marine diesel generator sets as described in claim 8, characterized in that, Combining twin simulation algorithms for real-time simulation and real-time risk indicator monitoring and adjustment also includes: Based on the global candidate adjustment action parameter set combined with the twin simulation model constructed by the ship diesel generator set, after abnormal simulation adjustment, the global inference response dataset is obtained. Based on the global simulation response dataset and combined with a multi-objective optimization decision-making algorithm, the optimal set of adjustment actions is obtained with the goal of maximizing the preset comprehensive decision factor. The comprehensive decision factor is obtained by weighting the comprehensive impact score and the comprehensive performance score. The comprehensive performance score is obtained by weighting the vibration suppression rate, temperature drop amplitude, and state stability recovery time of real-time monitoring. The comprehensive impact score is obtained by weighting the execution time of the optimal set of adjustment actions, equipment wear and tear, and the probability of action execution risk.

10. The vibration isolation condition detection equipment for marine diesel generator sets as described in claim 9, characterized in that, Combining twin simulation algorithms for real-time simulation and real-time risk indicator monitoring and adjustment also includes: When the real-time risk indicator value is less than or equal to the preset risk indicator value threshold and at least one of the first feature value, the second feature value, and the third feature value is greater than the preset abnormality judgment threshold, the vibration isolation state is judged to be locally abnormal and an alarm is triggered. Based on the feature value set corresponding to the local anomaly in the vibration isolation state, the feature value magnitude is combined with the feature value-fault type-anomaly degree mapping table to obtain the anomaly degree corresponding to the current local anomaly type combination set. Based on the anomaly degree corresponding to the current set of local anomaly type combinations, and combined with the anomaly type-adjustment action type-adjustment degree mapping table, a set of local candidate adjustment action parameters is obtained. Based on the local candidate adjustment action parameter set, the local optimal adjustment action combination set is obtained by repeating the twin simulation and multi-objective optimization process. When the real-time risk indicator value is less than or equal to the preset risk indicator value threshold and the first feature value, the second feature value, and the third feature value are all less than the preset anomaly judgment feature, the changing trend of the first feature value, the second feature value, and the third feature value is monitored in real time according to the preset data interval. When the difference between at least one feature value and the corresponding preset anomaly judgment threshold is less than the preset risk warning threshold, an early warning operation is performed. At the same time, the warning level is determined according to the changing rate of the corresponding feature value, and a real-time warning of the corresponding level is performed.

Images