A method for identifying sub-health status of industrial pipe gallery equipment based on time-frequency domain fusion features

Abstract

This invention discloses a method for identifying sub-health conditions of industrial pipe gallery equipment based on time-frequency domain fusion features, belonging to the field of equipment condition monitoring technology. The method includes: deploying accelerometers at the bearing housing and support connection points of the target equipment to simultaneously collect vibration signals from the target equipment and adjacent equipment; generating a structural propagation delay sequence based on the support connection path length and the longitudinal wave velocity of the material; performing a short-time Fourier transform on the vibration signals and aligning the time-frequency matrices of adjacent equipment with delay; calculating the local coherence coefficient and attenuating the propagation interference region to obtain the target source purification time-frequency matrix; and extracting the frequency shift, residual impact energy, and broadband noise rise to determine the sub-health state. This invention can reduce vibration propagation interference from adjacent equipment and improve the accuracy of early sub-health identification of industrial pipe gallery equipment.

Description

Technical Field

[0001] This invention relates to the field of equipment condition monitoring technology, and in particular to a method for identifying the sub-health status of industrial pipe gallery equipment based on time-frequency domain fusion features. Background Technology

[0002] As industrial utility tunnels develop towards long-distance, multi-equipment, and continuous operation, pumps, fans, compressors, conveying mechanisms, and their support structures are typically located within the same tunnel space and under the same load-bearing system. Equipment vibration signals not only include the operating status information of the target equipment's own bearings, rotors, couplings, and other components, but also superimposed structurally coupled vibration components propagated from adjacent equipment via supports, pipe racks, or foundations. Existing equipment condition identification technologies primarily focus on feature extraction from single-point vibration, sound, temperature, and current operating signals, and use time-domain statistics, frequency-domain spectral peaks, time-frequency images, or machine learning models to determine normal, abnormal, or fault states. These technologies have been widely applied in identifying typical faults. However, the sub-healthy state of industrial utility tunnel equipment often manifests as early degradation characteristics such as slow shift of the frequency sideband, weak residual impact energy, and gradual increase in broadband noise. These characteristics are small in amplitude and long in duration, easily masked by interference from adjacent equipment and background structural noise. If conventional time-frequency analysis is performed based solely on the target measurement point signal, it is difficult to distinguish between the source signal of the target device and the coupled propagation signal of adjacent devices, resulting in problems such as insufficient sensitivity, high false judgment rate, and insufficient interpretability of early degradation characteristics in the identification of sub-health status.

[0003] CN116092525B discloses a method for electrical equipment status sound recognition considering time-frequency domain feature fusion. This method acquires and preprocesses time-domain sound data, converts the time-domain signal to a frequency-domain signal using Fourier transform, and then employs embedding, packaging, and mutual information methods for frequency-domain feature selection, time-domain feature extraction, and time-frequency feature fusion to establish a neural network for recognizing three states: normal operation, normal leakage, and abnormal leakage. While this method can improve sound status recognition accuracy by leveraging the complementarity of time-domain and frequency-domain information, it primarily processes electrical equipment sound signals, focusing on feature dimensionality reduction and classification. It does not model the structural propagation delay, vibration coupling between adjacent equipment, and propagation interference regions caused by the shared support arrangement of multiple devices in industrial pipe corridors. Furthermore, it does not perform delay alignment of the time-frequency matrices of adjacent devices based on the support connection path length and material longitudinal wave velocity. Therefore, it is difficult to apply to target source purification recognition scenarios in the early, weakly degraded state of industrial pipe corridor equipment.

[0004] CN120524088A discloses a reinforcement learning-based method for identifying sub-health states in refrigeration plant operations. This method learns the complex relationships between operating parameters of refrigeration plant equipment through data acquisition, data preprocessing, reinforcement learning model construction, hyperparameter optimization, model training, and real-time monitoring and early warning. After identifying sub-health states, it pushes status information, cause analysis, and maintenance feedback. This method is suitable for comprehensive monitoring and closed-loop maintenance of refrigeration plant operating parameters. However, it focuses on strategy learning and early warning management between operating parameters and does not construct a propagation interference reduction mechanism based on the time-frequency local coherence relationship in industrial pipe gallery vibration signals. It also does not extract specific features for early degradation of rotating equipment, such as frequency shift, residual impact energy, and broadband noise rise, from the purified target source time-frequency matrix. Therefore, under conditions of crosstalk between multiple devices, its ability to distinguish the source of sub-health states and determine continuous deviations in the target equipment remains insufficient.

[0005] CN122067773A discloses a sub-health status assessment modeling method using terahertz multimodal data. This method simultaneously acquires terahertz time-domain spectral data, terahertz imaging data, and physiological sign data, and outputs the sub-health risk level after denoising, normalization, time alignment, and attention fusion. While this method emphasizes multimodal data fusion assessment, its application objects and signal mechanisms differ, and it does not address the coherent interference caused by the propagation of vibrations from industrial pipe gallery equipment along the support structure.

[0006] In summary, existing time-frequency domain fusion identification technologies and sub-health state assessment technologies still suffer from several drawbacks, including insufficient consideration of the structural coupling propagation of multiple devices in industrial pipe corridors, difficulty in separating propagation interference from adjacent devices from the vibration signals of the target device, and insufficient targeting of early weak degradation feature extraction. This invention provides a sub-health state identification method for industrial pipe corridor equipment based on time-frequency domain fusion features. It calculates the structural propagation delay sequence using the set of support connection path lengths and the longitudinal wave velocity of the support material, aligns the time-frequency matrices of adjacent devices with the delay, and attenuates the amplitude of the propagation interference region in the target device's time-frequency matrix based on the local coherence coefficient, forming a target source purification time-frequency matrix. Finally, it determines the sub-health state by combining the frequency conversion sideband drift, residual impact energy, broadband noise rise, and continuous deviation from the historical health baseline feature vector. This addresses the problem of insufficient accuracy in identifying target source purification and early sub-health states of industrial pipe corridor equipment under multi-source vibration coupling environments. Summary of the Invention

[0007] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0008] In view of the problems that existing technologies for identifying the sub-health status of industrial pipe gallery equipment have, such as the difficulty in separating the vibration propagation interference of adjacent equipment and the easy masking of the early weak degradation characteristics of the target equipment by structural coupling noise, this invention is proposed.

[0009] Therefore, the problem to be solved by this invention is how to identify and weaken the propagation interference components in the vibration signal of the target equipment in a multi-equipment co-support operation environment, and accurately extract the time-frequency domain fusion features reflecting the sub-health state of the target equipment.

[0010] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a method for identifying the sub-health state of industrial pipe gallery equipment based on time-frequency domain fusion features, which includes: Accelerometers are installed at the bearing housing of the target equipment and at the connection points of the brackets from each adjacent equipment to the target equipment. Vibration signals of the target equipment and adjacent equipment are collected synchronously, and the set of bracket connection path lengths is recorded. Based on the set of support connection path lengths and the longitudinal wave velocity of the support material, the structural propagation delay time of vibration signals from each adjacent device to the target device measurement point is calculated, and a structural propagation delay sequence is generated. Short-time Fourier transforms are performed on the vibration signals of the target equipment and the adjacent equipment respectively to obtain the time-frequency matrix of the target equipment and the time-frequency matrix of the adjacent equipment. The time-frequency matrices of the adjacent equipment are then delayed and aligned according to the structure propagation delay sequence. Calculate the local coherence coefficients of the target device's time-frequency matrix and the time-frequency matrices of adjacent devices after delay alignment, and perform amplitude attenuation on the propagation interference region in the target device's time-frequency matrix according to the local coherence coefficients to obtain the target source purification time-frequency matrix; Extract the frequency shift, residual impact energy, and broadband noise rise from the target source purification time-frequency matrix to construct a sub-health coherent feature vector and determine the sub-health status of the target equipment.

[0011] Compared with existing technologies, the beneficial effects of this invention are as follows: By synchronously collecting vibration signals of the target equipment and adjacent equipment and recording the length of the support connection path, the source and propagation path of vibration interference from adjacent equipment are clarified; by calculating the structural propagation delay time and generating a structural propagation delay sequence, the vibration signals of adjacent equipment can be accurately aligned with the vibration signals of the target equipment in the time-frequency domain; by constructing a time-frequency matrix through short-time Fourier transform and identifying the propagation interference region based on the local coherence coefficient, the amplitude of the interference component in the time-frequency matrix of the target equipment is attenuated to obtain a target source purification time-frequency matrix that better reflects the state of the target equipment itself; furthermore, the frequency conversion sideband drift, impact energy residue, and broadband noise rise are extracted from the target source purification time-frequency matrix to construct a sub-health coherent feature vector, and the comprehensive deviation of the continuous monitoring window is used for judgment, thereby reducing misjudgment caused by coupled vibration of adjacent equipment and improving the accuracy, stability, and interpretability of identifying the sub-health state of industrial pipe gallery equipment. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart of a method for identifying the sub-health status of industrial pipe gallery equipment based on time-frequency domain fusion features. Detailed Implementation

[0013] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0014] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

[0015] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0016] As mentioned in the background section, while existing technologies can perform time-frequency analysis or sub-health status assessment of equipment operating signals, they typically do not address vibration interference from adjacent equipment by considering factors such as support connection paths, structural propagation delays, and local coherence relationships. This results in insufficient accuracy and stability in identifying the sub-health status of the target equipment. To address these issues, this invention provides a method for identifying the sub-health status of industrial pipe gallery equipment based on time-frequency domain fusion features.

[0017] Reference Figure 1 , Figure 1 This is a flowchart illustrating a method for identifying the sub-health state of industrial pipe gallery equipment based on time-frequency domain fusion features, according to an embodiment of the present invention. Figure 1 As shown, a method for identifying the sub-health state of industrial pipe gallery equipment based on time-frequency domain fusion features includes: S1: Install acceleration sensors at the bearing housing of the target equipment and at the connection points of the brackets from each adjacent equipment to the target equipment to synchronously collect vibration signals of the target equipment and adjacent equipment, and record the set of bracket connection path lengths.

[0018] S1.1: Using the equipment to be identified in the industrial pipe gallery as the target equipment, read the equipment number, installation section and equipment type of the target equipment, and determine the equipment that shares a bracket with the target equipment or is continuously connected through the bracket based on the installation section, and record it as adjacent equipment.

[0019] S1.1.1: Read the equipment number, installation section, equipment type and corresponding bracket number of the target equipment in the industrial pipe gallery equipment ledger, and mark the location of the bearing seat of the target equipment as the target equipment measuring point.

[0020] It should be noted that the target device measurement point is the location where the vibration signal of the target device is subsequently collected, and it is also the reference point for calculating the arrival position of the vibration signal of each adjacent device.

[0021] S1.1.2: Using the installation section of the target equipment as the search scope, query equipment that has the same bracket number, continuous bracket number or shared beam number as the target equipment to obtain a set of candidate adjacent equipment.

[0022] Specifically, the candidate set of adjacent equipment includes fans, pumps, motors, or valve actuators that have a support connection with the target equipment.

[0023] S1.1.3: Perform a support continuity check on each candidate adjacent device in the candidate adjacent device set. Record the candidate adjacent devices that can be continuously connected to the target device measuring point via the pipe gallery support as adjacent devices, and obtain the adjacent device list.

[0024] Furthermore, if there is a disconnected support section, flexible vibration isolation section, or non-load-bearing connection section between a candidate adjacent device and the target device's measuring point, then the candidate adjacent device will not be included in the adjacent device list.

[0025] In an optional embodiment, when the target device is a drainage pump, if the adjacent fan and the adjacent motor are both installed on the crossbeam of the pipe gallery that is continuously connected to the drainage pump, then the adjacent fan and the adjacent motor are recorded in the adjacent device list; if the adjacent distribution cabinet and the drainage pump are only connected by a cable tray and there is no continuous support connection, then the adjacent distribution cabinet is not recorded in the adjacent device list.

[0026] S1.2: Install the target acceleration sensor on the bearing housing of the target equipment, and use the installation point of the target acceleration sensor as the measurement point of the target equipment.

[0027] Furthermore, the sensitive axis of the target accelerometer is aligned with the main vibration direction of the target equipment bearing housing in order to obtain the vibration signal of the target equipment.

[0028] S1.3: Install adjacent acceleration sensors at the bracket connection points from each adjacent device to the target device, and assign adjacent device numbers to each adjacent acceleration sensor.

[0029] S1.3.1: According to the list of adjacent equipment, determine the continuous connection path of the support between each adjacent equipment and the target equipment measuring point, and select the support connection point on the side closest to the corresponding adjacent equipment in each continuous connection path. The support connection point includes support node, beam node or fixed clamping node, and the support connection point is located on the transmission path of the vibration signal of the adjacent equipment into the pipe gallery support.

[0030] S1.3.2: Install adjacent accelerometers at the connection points of each support, and ensure that the sensitive axis direction of each adjacent accelerometer is consistent with the extension direction of the corresponding support continuous connection path.

[0031] It should be noted that each adjacent acceleration sensor is used to collect the vibration signal of the adjacent device transmitted through the bracket.

[0032] S1.3.3: Configure adjacent device numbers for each adjacent accelerometer and bind the adjacent device numbers to the corresponding adjacent devices and the installation locations of adjacent accelerometers.

[0033] Preferably, the adjacent device numbers are consistent in the subsequent set of support connection path lengths, the structure propagation delay sequence, and the adjacent device time-frequency matrix.

[0034] S1.4: Measure the length of the support connection path from each adjacent device to the target device measuring point along the continuous connection direction of the support corresponding to the number of each adjacent device.

[0035] S1.4.1: Using the installation position of each adjacent acceleration sensor as the starting point of the path and the target device measurement point as the ending point of the path, determine the support measurement path corresponding to each adjacent device number. The support measurement path is determined along the actual connection direction of the pipe gallery support.

[0036] S1.4.2: Divide the measurement path of each support into several support segments according to the adjacent support nodes, and measure the support segment length of each support segment. The support segment length is the length between two adjacent support nodes along the support centerline.

[0037] S1.4.3: Add up the lengths of all bracket segments corresponding to the same adjacent equipment number to obtain the bracket connection path length corresponding to that adjacent equipment number.

[0038] It should be noted that the length of the support connection path is not the straight-line distance between the installation positions of adjacent accelerometers and the measuring points of the target device.

[0039] In an optional embodiment, if the first adjacent device passes through the first support segment, the second support segment, and the third support segment in sequence from the target device measurement point, and the lengths of the three segments are 2.0m, 1.5m, and 3.0m respectively, then the support connection path length corresponding to the first adjacent device number is 6.5m.

[0040] S1.5: Bind each adjacent device number to the corresponding bracket connection path length, and arrange them in the order of adjacent device numbers to obtain the bracket connection path length set.

[0041] It should be noted that each item in the set of support connection path lengths corresponds to an adjacent device number, which is used for subsequent calculation of structural propagation delay time.

[0042] S1.6: Configure the target accelerometer and each adjacent accelerometer with the same sampling frequency and the same acquisition start time, and acquire the vibration signals of the target equipment and the adjacent equipment respectively.

[0043] S1.6.1: Configure the target accelerometer and all adjacent accelerometers with the same sampling frequency, the same acquisition start time, and the same monitoring window length.

[0044] Preferably, the sampling frequency is not less than ten times the sampling frequency corresponding to the rated rotation frequency of the target device.

[0045] S1.6.2: Start the target acceleration sensor and each adjacent acceleration sensor according to the start time of data acquisition, and collect the vibration signals of the target equipment and the adjacent equipment.

[0046] Furthermore, both the vibration signals of the target device and the vibration signals of adjacent devices are accompanied by a timestamp of the acquisition.

[0047] S1.6.3: Time-stamp the vibration signals of the target device and adjacent devices according to the acquisition timestamp, and arrange the vibration signals of adjacent devices according to the adjacent device number.

[0048] It should be noted that the order of vibration signals from adjacent devices is consistent with the order of adjacent device numbers in the set of support connection path lengths.

[0049] S2: Based on the set of support connection path lengths and the longitudinal wave velocity of the support material, calculate the structural propagation delay time of vibration signals from each adjacent device to the target device measuring point, and generate a structural propagation delay sequence.

[0050] S2.1: Read the set of support connection path lengths and call the longitudinal wave velocity of the support material corresponding to the installation section of the target equipment.

[0051] Furthermore, the installation section of the target equipment is read, and the pipe rack support ledger is consulted based on the installation section to determine the corresponding pipe rack support material type for the installation section of the target equipment. The pipe rack support material type includes carbon steel support, stainless steel support, or composite support. Based on the pipe rack support material type, the corresponding support material longitudinal wave velocity is read from the preset support material parameter table. The support material longitudinal wave velocity is the velocity parameter used when vibration propagates along the pipe rack support material, not the airborne sound propagation velocity. The support material longitudinal wave velocity is bound to the installation section of the target equipment as a unified velocity parameter for subsequent calculations of the vibration signals of each adjacent device reaching the target equipment measurement point. When the support material type is consistent within the installation section of the target equipment, the longitudinal wave velocity of the same support material is used for the length of all support connection paths.

[0052] S2.2: According to the adjacent equipment number, read the corresponding support connection path length from the support connection path length set item by item, where the support connection path length is the actual path length of the vibration signal of the adjacent equipment transmitted to the target equipment measuring point along the continuous connection direction of the support.

[0053] S2.3: Calculate the ratio of the length of each support connection path to the longitudinal wave velocity of the support material to obtain the structural propagation delay time of the vibration signal of each adjacent device to the target device measuring point.

[0054] Specifically, according to the adjacent equipment number, the corresponding support connection path length and longitudinal wave velocity of the support material are read one by one; the support connection path length of each item is divided by the longitudinal wave velocity of the support material to obtain the structural propagation delay time of the corresponding adjacent equipment number, where the structural propagation delay time represents the time required for the vibration signal of the corresponding adjacent equipment to be transmitted to the target equipment measuring point through the pipe gallery support; each structural propagation delay time is bound with the corresponding adjacent equipment number to obtain the initial structural propagation delay data.

[0055] It should be noted that each item in the initial structure propagation delay data comes from the length of a single support connection path, and path data corresponding to different adjacent device numbers are not mixed.

[0056] In an optional embodiment, if the length of the support connection path corresponding to the first adjacent device is 5.9m and the longitudinal wave velocity of the support material is 5900m / s, then the structural propagation delay time corresponding to the first adjacent device is 0.001s.

[0057] S2.4: Based on the sampling frequency of the vibration signal of the target equipment and the vibration signal of the adjacent equipment, the propagation delay time of each structure is converted into the corresponding sampling point delay number.

[0058] Furthermore, the sampling point delay number serves as the basis for subsequent displacement of the time axis delay alignment of the time-frequency matrices of adjacent devices.

[0059] The sampling frequencies used for the vibration signals of the target device and adjacent devices are read, with the target device vibration signal and adjacent device vibration signals using the same sampling frequency. The propagation delay time of each structure is multiplied by the sampling frequency to obtain the sampling point delay number corresponding to the adjacent device number. When the sampling point delay number has a decimal part, the integer part is recorded as the time axis translation point number, and the fractional sampling delay corresponding to the decimal part is recorded as the intra-frame compensation delay (unit: sampling period) for subsequent phase compensation.

[0060] It should be noted that the sampling point delay number represents the number of delay points of the vibration signal of the corresponding adjacent device relative to the vibration signal of the target device in the sampling sequence; the number of time axis translation points and the intra-frame compensation delay together serve as the basis for the subsequent time-frequency matrix delay alignment of adjacent devices.

[0061] In an optional embodiment, if the structural propagation delay time corresponding to the first adjacent device is 0.001s and the sampling frequency is 10000Hz, then the number of sampling point delays corresponding to the first adjacent device is 10 sampling points (integer, without decimal part).

[0062] S2.5: Establish a correspondence between the numbers of each adjacent device, the propagation delay time of each structure, and the delay number of each sampling point to form a structure propagation delay record.

[0063] Furthermore, using adjacent device numbers as indexes, the corresponding support connection path length, structural propagation delay time, and sampling point delay number are written into the same structural propagation delay record; consistency checks are performed on each structural propagation delay record, including whether adjacent device numbers are duplicated, whether the support connection path length is empty, and whether the structural propagation delay time is greater than zero; the structural propagation delay records after consistency checks are then entered into the subsequent sorting process.

[0064] S2.6: Sort all structure propagation delay records according to the order of adjacent device numbers to generate a structure propagation delay sequence.

[0065] Preferably, the structure propagation delay sequence adopts the adjacent device numbering order that is the same as the set of support connection path lengths.

[0066] S3: Perform short-time Fourier transform on the vibration signals of the target equipment and the adjacent equipment respectively to obtain the time-frequency matrix of the target equipment and the time-frequency matrix of the adjacent equipment, and perform delay alignment on the time-frequency matrix of the adjacent equipment according to the structure propagation delay sequence.

[0067] S3.1: Read the vibration signal of the target device, the vibration signal of the adjacent device, and the structural propagation delay sequence, wherein the structural propagation delay sequence includes the structural propagation delay time and the number of sampling point delays corresponding to each adjacent device number.

[0068] S3.2: Configure the same short-time Fourier transform parameter set for the vibration signal of the target equipment and the vibration signal of the adjacent equipment.

[0069] S3.2.1: Read the sampling frequency of the vibration signal of the target equipment and the rated rotation frequency of the target equipment, and determine the frequency resolution requirement based on the rated rotation frequency. The frequency resolution requirement is the minimum frequency interval that can distinguish adjacent rotation frequency sideband components.

[0070] Specifically, the frequency resolution requirement is one-tenth to one-fifth of the rated rotation frequency to ensure that there are at least two frequency units between adjacent rotation frequency sideband components in the time-frequency matrix of the target device; if the rated rotation frequency of the target device is 50Hz, then the frequency resolution requirement is 5Hz to 10Hz.

[0071] S3.2.2: Based on the frequency resolution requirement and the sampling frequency, determine the window length, where the window length is the number of sampling points corresponding to the rounded ratio of the sampling frequency to the frequency resolution requirement.

[0072] It should be noted that the window length determines the frequency resolution of the short-time Fourier transform. The larger the window length, the higher the frequency resolution, but the coarser the time-varying features captured within the same time window. The smaller the window length, the higher the time resolution, but the lower the frequency resolution. Determining the window length should prioritize meeting the frequency resolution requirements.

[0073] Preferably, the window length is taken as an integer power of two so that the number of subsequent Fourier transform points is consistent with the window length, and the computational load is reduced by using the fast Fourier transform algorithm. When there is a significant speed fluctuation in the target device, the window length is adjusted in real time according to the current speed to keep the number of frequency cycles contained in each time window constant within the range of 8 to 16 cycles.

[0074] In an optional embodiment, when the sampling frequency is 10000Hz and the frequency resolution requirement is 5Hz, the window length is 10000÷5=2000 sampling points, which is adjusted to 2048 sampling points after being raised to the power of two.

[0075] S3.2.3: Determine the window shift length based on the window length, where the window shift length is the sampling point interval between the starting points of adjacent time windows.

[0076] Furthermore, the window shift length is one-quarter to one-half of the window length; preferably, the window shift length is one-half of the window length, corresponding to an overlap rate of 50% between adjacent time windows, achieving a balance between time resolution and computational load; the ratio of the window shift length to the window length is denoted as the window shift ratio, which is used to convert the number of sampling point delays into the time window offset.

[0077] S3.2.4: Select the window function type, and denote the window function type, window length, window shift length, and number of Fourier transform points as the short-time Fourier transform parameter set. The short-time Fourier transform parameter set should be consistent with the vibration signals of the target equipment and all adjacent equipment.

[0078] It should be noted that the window function type selected is the Hanning window; the short-time Fourier transform parameter set is consistent with the vibration signals of the target device and all adjacent devices to ensure that the subsequent time-frequency matrix of the target device and the time-frequency matrix of each adjacent device are completely corresponding in terms of frequency axis unit division and time axis window division, so that the time-frequency unit calculation of the local coherence coefficient is performed between the dimensionally matched matrices.

[0079] S3.3: According to the short-time Fourier transform parameter set, the vibration signal of the target device is subjected to frame windowing and short-time Fourier transform to obtain the time-frequency matrix of the target device, where the rows of the time-frequency matrix of the target device correspond to the frequency unit and the columns correspond to the time window.

[0080] S3.4: According to the short-time Fourier transform parameter set, perform frame windowing and short-time Fourier transform on the vibration signals of each adjacent device to obtain the time-frequency matrix of each adjacent device.

[0081] It should be noted that the time-frequency matrices of each adjacent device and the target device use the same frequency unit and time window arrangement.

[0082] S3.5: Based on the structure propagation delay sequence, read the sampling point delay number corresponding to each adjacent device number, and calculate the time window offset corresponding to the time frequency matrix of each adjacent device according to the sampling point delay number and the window shift length.

[0083] S3.5.1: According to the adjacent device number, read the sampling point delay number corresponding to each adjacent device number from the structure propagation delay sequence one by one. The sampling point delay number includes the number of time axis translation points (integer part) and the intra-frame compensation delay (the number of sampling periods corresponding to the fractional part).

[0084] It should be noted that the number of time axis translation points is an integer number of sampling points, and the intra-frame compensation delay is a fractional delay of less than one sampling period. Both are used in steps S3.5 and S3.6 for delay alignment operations at different precision levels.

[0085] S3.5.2: Divide the number of time axis translation points corresponding to each adjacent device number by the determined window shift length, take the integer part of the quotient as the time window offset corresponding to that adjacent device number, and take the remainder as the number of intra-frame remainder points.

[0086] Furthermore, the time window offset represents the integer number of windows that the time-frequency matrix of the corresponding adjacent device needs to be shifted backward along the time axis; the sum of the intra-frame margin points and the intra-frame compensation delay (in units of sampling period) is denoted as the comprehensive intra-frame margin points (which can be non-integer), and is used for phase compensation.

[0087] In an optional embodiment, if the number of time axis translation points corresponding to the first adjacent device is 110 sampling points and the window shift length is 1024 sampling points, then the time window offset is 0 windows and the number of intra-frame reserve points is 110 sampling points; if the number of time axis translation points corresponding to the second adjacent device is 2150 sampling points and the window shift length is 1024 sampling points, then the time window offset is 2 windows and the number of intra-frame reserve points is 102 sampling points.

[0088] S3.5.3: Bind each adjacent device number with its corresponding time window offset and the total number of intra-frame margin points to obtain a time alignment parameter table.

[0089] It should be noted that the order of adjacent device numbers in the time alignment parameter table is consistent with the order of adjacent device numbers in the structure propagation delay sequence.

[0090] S3.6: According to the offset of each time window, the time-frequency matrix of the corresponding adjacent device is delayed and aligned along the time axis to obtain the aligned time-frequency matrix of each adjacent device.

[0091] S3.6.1: According to the adjacent device number, read the time window offset corresponding to the first adjacent device number from the time alignment parameter table, shift the adjacent device time-frequency matrix corresponding to the adjacent device number backward along the time axis by the time window offset column, add zero value columns equal to the time window offset to the beginning of the shifted matrix, and truncate the part that exceeds the original number of columns at the end to obtain the full window aligned time-frequency matrix corresponding to the adjacent device number.

[0092] It should be noted that the number of columns in the full-window aligned time-frequency matrix is ​​the same as that in the target device's time-frequency matrix; when the time window offset is zero, the adjacent device's time-frequency matrix does not undergo column shifting and directly proceeds to S3.6.2; the added zero-value column indicates that within the time range corresponding to the vibration signal propagation delay of the adjacent device, the adjacent device has not yet contributed to the vibration propagation of the target device's measuring point.

[0093] S3.6.2: Read the number of integrated intra-frame margin points corresponding to the adjacent device number from the time alignment parameter table. When the number of integrated intra-frame margin points is not zero, perform intra-frame compensation on the phase of each time-frequency unit in the whole window alignment time-frequency matrix according to the number of integrated intra-frame margin points to obtain the phase-compensated time-frequency matrix.

[0094] Specifically, intra-frame phase compensation is achieved by aligning the complex values ​​of each frequency element in the time-frequency matrix along the frequency axis to the entire window and multiplying them by a phase rotation factor. ,in The imaginary unit, The center frequency of the k-th frequency unit. The phase rotation factor is determined by the center frequency of each frequency unit and the time length corresponding to the number of residual points in the overall frame. The time length corresponding to the number of residual points in the overall frame is the number of residual points in the overall frame divided by the sampling frequency.

[0095] Preferably, the rotation factor corresponds to the forward translation Δt operation in the time domain, which is used to accurately compensate for the sub-window-level time delay deviation remaining after the integer window translation, so that the vibration signal of the adjacent device and the vibration signal of the target device can achieve synchronization at the sampling point level in the time domain; when the total number of residual points in the frame is zero, the whole window aligned time-frequency matrix is ​​directly used as the phase compensation time-frequency matrix.

[0096] It should be noted that intra-frame phase compensation corrects the sub-window level time delay deviation introduced by the overall intra-frame margin point count. If the overall intra-frame margin point count is small, the effect of intra-frame phase compensation on low-frequency components can be ignored, but the phase correction effect on high-frequency components is significant. If the sub-health characteristics of the target device are mainly distributed in the low-frequency sub-band, intra-frame phase compensation can be omitted, and the whole window aligned time-frequency matrix can be directly used as the phase compensation time-frequency matrix.

[0097] In an optional embodiment, if the number of integrated intra-frame margin points corresponding to the second adjacent device is 102 sampling points and the sampling frequency is 10000Hz, then Δt = 102 / 10000 = 0.0102 s; for the component with a center frequency of 100Hz, the phase rotation factor corresponds to a phase angle of 2π × 100 × 0.0102 = 6.408 rad, and the complex value is multiplied by... For the component with a center frequency of 1000Hz, the phase rotation factor corresponds to a phase angle of 2π×1000×0.0102=64.09 rad, and the phase correction effect is significant.

[0098] S3.6.3: Record the phase compensation time-frequency matrix corresponding to each adjacent device number as the adjacent device alignment time-frequency matrix corresponding to that adjacent device number, arrange them in order of adjacent device number, and obtain the set of all adjacent device alignment time-frequency matrices.

[0099] S4: Calculate the local coherence coefficients of the target device's time-frequency matrix and the time-frequency matrices of adjacent devices after delay alignment. Attenuate the amplitude of the propagation interference region in the target device's time-frequency matrix according to the local coherence coefficients to obtain the target source purification time-frequency matrix.

[0100] S4.1: Read the target device's time-frequency matrix and the time-frequency matrices of each adjacent device, and confirm that the target device's time-frequency matrix and the time-frequency matrices of each adjacent device have the same frequency units and time windows.

[0101] Specifically, the target device's time-frequency matrix and the time-frequency matrices aligned with each adjacent device are both complex matrices, containing amplitude and phase information.

[0102] S4.2: Using each time-frequency unit in the target device's time-frequency matrix as the center, select a preset number of adjacent frequency units and adjacent time windows to construct a local time-frequency neighborhood.

[0103] S4.2.1: Read the short-time Fourier transform parameter set, extract the window length and window shift length, calculate the frequency unit interval of the target device time-frequency matrix based on the window length and sampling frequency, and calculate the time window interval of the target device time-frequency matrix based on the window shift length and sampling frequency.

[0104] S4.2.2: Based on the rated rotation frequency of the target device, determine the frequency axis neighborhood radius of the local time-frequency neighborhood. The frequency axis neighborhood radius is the number of frequency units that can cover a range of one times the rated rotation frequency. That is, the frequency axis neighborhood radius is the number of frequency units obtained by dividing the rated rotation frequency by the frequency unit interval and then rounding it down.

[0105] It should be noted that the determination of the frequency axis neighborhood radius is based on the fact that the propagation interference components of the rotating equipment in the pipe gallery diffuse outwards in both directions in the time-frequency matrix with the frequency as the basic unit. The frequency axis neighborhood radius covering one frequency range can completely capture the coherent energy diffusion caused by frequency modulation. When the rated frequency of the target equipment is 25Hz and the frequency unit interval is 5Hz, the frequency axis neighborhood radius is 25÷5=5 frequency units.

[0106] S4.2.3: Determine the time axis neighborhood radius of the local time-frequency neighborhood based on the time stability of the sub-health characteristics of the target device.

[0107] Furthermore, the time axis neighborhood radius is set to three to five time windows. If the time axis neighborhood radius is set to three time windows, then a total of seven time windows, centered on the current time-frequency unit and encompassing the range of three time windows before and after it, will be used to calculate the local coherence coefficient. The value of the time axis neighborhood radius should not be too large. When the time axis neighborhood radius exceeds seven time windows, the time range spanned by the local time-frequency neighborhood becomes too long, causing non-steady-state characteristics introduced by changes in equipment operating conditions to be mixed into the local time-frequency neighborhood, thus reducing the accuracy of the local coherence coefficient in identifying propagating interference components.

[0108] Preferably, both the time axis neighborhood radius and the frequency axis neighborhood radius are odd numbers, and are symmetrically distributed around the calculated time-frequency unit to avoid asymmetrical truncation of the local time-frequency neighborhood in the time axis or frequency axis direction.

[0109] S4.2.4: Using the radius of the frequency axis neighborhood and the radius of the time axis neighborhood as boundaries, construct a rectangular local time-frequency neighborhood centered on each time-frequency unit in the time-frequency matrix of the target device.

[0110] Furthermore, when the time-frequency unit is located at the edge of the target device's time-frequency matrix, causing the local time-frequency neighborhood to exceed the matrix boundary, the portion exceeding the boundary is filled with zero values. The number of elements in the frequency axis direction of the rectangular local time-frequency neighborhood is twice the radius of the frequency axis neighborhood plus one, and the number of elements in the time axis direction is twice the radius of the time axis neighborhood plus one.

[0111] In an optional embodiment, when the frequency axis neighborhood radius is 5 frequency units and the time axis neighborhood radius is 3 time windows, the rectangular local time-frequency neighborhood corresponding to each time-frequency unit contains 11×7 total 77 time-frequency units. The target device time-frequency matrix and the time-frequency matrix aligned with each adjacent device each provide 77 complex amplitude samples in the same rectangular local time-frequency neighborhood for the calculation of cross power spectrum and self power spectrum.

[0112] S4.3: Within each local time-frequency neighborhood, calculate the cross power spectrum and self power spectrum between the time-frequency matrix of the target device and the aligned time-frequency matrix of the adjacent device.

[0113] Specifically, the cross-power spectrum is calculated by the conjugate product of the target device's time-frequency matrix and the aligned time-frequency matrix of the adjacent device, while the self-power spectrum is calculated by the square of the magnitude of each matrix.

[0114] S4.4: Based on the cross power spectrum and the self power spectrum, calculate the local coherence coefficients of the target device's time-frequency matrix and the time-frequency matrix aligned with each neighboring device in each local time-frequency neighborhood.

[0115] Furthermore, the local coherence coefficient is used to characterize the degree of correlation between the vibration signals of adjacent devices and the propagation of the target device's measuring point in the same time-frequency region.

[0116] S4.5: For the same time-frequency unit, compare the local coherence coefficients corresponding to the aligned time-frequency matrices of each adjacent device, and select the largest local coherence coefficient as the propagation interference determination coefficient of the time-frequency unit.

[0117] It should be noted that the adjacent device number record corresponding to the propagation interference determination coefficient is the device number of the interference source.

[0118] S4.6: Compare the propagation interference determination coefficient with the preset coherence threshold. If the propagation interference determination coefficient is greater than the preset coherence threshold, then this frequency unit is marked as a propagation interference unit, and all propagation interference units constitute the propagation interference region. If the propagation interference determination coefficient is less than or equal to the preset coherence threshold, then this frequency unit is marked as a non-propagation interference unit.

[0119] S4.6.1: Before the target equipment is put into formal monitoring, under the baseline operating condition where the target equipment and all adjacent equipment are in normal operation, collect the local coherence coefficients of no less than 30 monitoring windows according to the operation of S4.1 to S4.5. Take the mean and standard deviation of the local coherence coefficients at each time-frequency unit within the 30 monitoring windows to obtain the baseline coherence mean and baseline coherence standard deviation corresponding to each time-frequency unit.

[0120] It should be noted that the baseline operating condition is the confirmed normal operating stage of the target equipment and all adjacent equipment excluding sub-health conditions. The baseline coherence mean and baseline coherence standard deviation reflect the statistical distribution of the propagation coherence characteristics at each time-frequency unit during the healthy operating stage, and serve as the basis for setting the preset coherence threshold. The acquisition time span of the 30 monitoring windows is no less than 72 hours to cover the coherence coefficient distribution changes of the target equipment under different load conditions.

[0121] S4.6.2: The preset coherence threshold of each time-frequency unit is the mean of the reference coherence plus twice the standard deviation of the reference coherence. The preset coherence thresholds of all time-frequency units are arranged into a preset coherence threshold matrix, wherein the preset coherence threshold matrix has the same row and column dimensions as the target device time-frequency matrix.

[0122] Furthermore, the preset coherence thresholds for different frequency units and different time windows in the preset coherence threshold matrix are different. For time-frequency units corresponding to frequency sub-bands where propagation interference energy is concentrated, the reference coherence standard deviation is small, and the preset coherence threshold is close to the upper limit of the reference coherence mean plus twice the reference coherence standard deviation. For time-frequency units corresponding to frequency sub-bands dominated by background noise, the reference coherence standard deviation is large, and the preset coherence threshold is increased accordingly to reduce the misjudgment rate of background noise on the labeling results of propagation interference units.

[0123] Preferably, the preset coherence threshold matrix is ​​updated every preset update cycle based on the latest collected benchmark operating condition data, with the preset update cycle being no less than 30 days, to reflect the drift of the coherence characteristic baseline caused by bracket aging or changes in sensor installation status; when sufficient benchmark operating condition data cannot be obtained, all elements in the preset coherence threshold matrix are assigned a uniform value of 0.7.

[0124] S4.6.3: Based on the above comparison, all time-frequency units in the target device's time-frequency matrix are divided into propagation interference regions and non-propagation interference regions.

[0125] It should be noted that the propagation interference area and the non-propagation interference area together cover all time-frequency units of the target device's time-frequency matrix, with no overlap or omissions; the amplitude of each time-frequency unit in the propagation interference area undergoes amplitude attenuation processing, while the amplitude of each time-frequency unit in the non-propagation interference area is directly retained in S4.8.

[0126] In an optional embodiment, if the target device time-frequency matrix contains 512 frequency units and 200 time windows, totaling 102,400 time-frequency units, after being marked in S4.6.3, 8,320 of these time-frequency units are marked as propagation interference units, constituting a propagation interference region, while the remaining 94,080 time-frequency units constitute a non-propagation interference region.

[0127] S4.7: Calculate the amplitude attenuation coefficient based on the propagation interference determination coefficient, and attenuate the amplitude of the propagation interference region in the time-frequency matrix of the target device according to the amplitude attenuation coefficient.

[0128] S4.7.1: For each propagation interference unit within the propagation interference area, read the propagation interference determination coefficient corresponding to the propagation interference unit from the propagation interference determination coefficient matrix, and calculate the amplitude attenuation coefficient corresponding to the propagation interference unit based on the propagation interference determination coefficient, wherein the amplitude attenuation coefficient is obtained by subtracting the propagation interference determination coefficient from one.

[0129] It should be noted that the amplitude attenuation coefficient ranges from [0,1]. When the propagation interference determination coefficient is one, the amplitude attenuation coefficient is zero, and the amplitude of the target device's time-frequency matrix at that time-frequency unit is completely set to zero, indicating that all the vibration energy at that time-frequency unit comes from the interference component propagated by the adjacent device through the support. When the propagation interference determination coefficient is equal to the preset coherence threshold, the amplitude attenuation coefficient is one minus the preset coherence threshold, and the amplitude at that time-frequency unit is partially retained to prevent the components in the target device's own degradation signal that have a weak correlation with the propagation interference component from being oversuppressed.

[0130] S4.7.2: Set a lower limit for the amplitude attenuation coefficient. When the amplitude attenuation coefficient is less than the lower limit, the amplitude attenuation coefficient of the time-frequency unit is taken as the lower limit.

[0131] Specifically, the lower limit is set to [0.1, 0.2], and the lower limit is set to 0.1 to ensure that any time-frequency unit in the propagation interference area retains a residual component of no less than 10% of the original amplitude after amplitude attenuation, preventing the target device's degradation characteristics from being completely erased due to the calculation error of the propagation interference determination coefficient. It should be noted that the setting of the lower limit is for the case where the amplitude attenuation coefficient in the propagation interference area is too small. It is not applicable to time-frequency units in non-propagation interference areas. The amplitude of time-frequency units in non-propagation interference areas is directly retained in S4.8 without the calculation of the amplitude attenuation coefficient.

[0132] S4.7.3: Multiply the complex value of each propagation interference element in the time-frequency matrix of the target device by the corresponding amplitude attenuation coefficient to obtain the attenuated complex value of each propagation interference element.

[0133] It should be noted that the amplitude of the attenuated complex value is the product of the original amplitude and the amplitude attenuation coefficient, and the phase remains consistent with the original phase. The amplitude attenuation process only applies to the amplitude of each time-frequency unit in the target device's time-frequency matrix, without changing the phase of the corresponding time-frequency unit. The phase information is used to determine the offset direction of the sideband component when extracting the frequency shift in S5.1. The characteristic of the attenuated complex value retaining the phase ensures the phase continuity of the target source purification time-frequency matrix in the time-frequency domain, thus giving physical meaning to the sub-healthy coherent feature vector extracted based on the target source purification time-frequency matrix in S5.

[0134] In an optional embodiment, if the propagation interference determination coefficient corresponding to a certain propagation interference unit is 0.85, then the amplitude attenuation coefficient is 1-0.85=0.15, which is greater than the lower limit of 0.1. After the amplitude attenuation, the propagation interference unit retains 15% of the original amplitude. If the propagation interference determination coefficient corresponding to another propagation interference unit is 0.95, then the amplitude attenuation coefficient is 1-0.95=0.05, which is less than the lower limit of 0.1. After the amplitude attenuation, the amplitude attenuation coefficient of the propagation interference unit takes the lower limit of 0.1, and after the amplitude attenuation, it retains 10% of the original amplitude.

[0135] S4.8: Retain the amplitude and phase of the non-propagation interference region in the target device's time-frequency matrix, and merge the propagation interference region after amplitude attenuation processing with the non-propagation interference region to obtain the target source purification time-frequency matrix.

[0136] Preferably, the target source purification time-frequency matrix reduces the synchronous interference components generated by adjacent equipment propagating along the pipe gallery support, while retaining the local impact and sideband drift components corresponding to the early sub-health of the target equipment.

[0137] S5: Extract the frequency shift, residual impact energy, and broadband noise rise from the target source purification time-frequency matrix to construct a sub-healthy coherent feature vector and determine the sub-healthy state of the target equipment.

[0138] S5.1: Read the target source purification time-frequency matrix and call the target device's rated frequency, historical health baseline parameter set, and preset deviation threshold.

[0139] Specifically, the historical health baseline parameter set includes: frequency switching sideband baseline value (center-of-gravity frequency of sideband energy in a healthy state), impulse energy baseline value (peak energy of high-frequency impulses in a healthy state), broadband noise baseline value (average energy of broadband noise in a healthy state), and corresponding standard deviations: frequency switching sideband baseline standard deviation, impulse energy baseline standard deviation, and broadband noise baseline standard deviation.

[0140] S5.2: Centered on the rated frequency, select the frequency-return neighborhood and the sideband neighborhood in the target source purification time-frequency matrix, extract the energy ridge line position within the frequency-return neighborhood and the sideband neighborhood, and take the offset of the energy ridge line position relative to the frequency-return sideband baseline value within the current monitoring window as the frequency-return sideband drift.

[0141] S5.2.1: Based on the rated frequency and the frequency unit interval of the target source purification time-frequency matrix, determine the frequency range of the frequency neighborhood. The frequency neighborhood is centered on the frequency unit corresponding to the rated frequency, and extends to both sides by one-fifth of the rated frequency corresponding to the frequency unit, thus obtaining the frequency range of the frequency neighborhood.

[0142] It should be noted that the frequency range of the frequency neighborhood is determined based on the fact that the frequency component of the rotating equipment in the pipe gallery exhibits a slight diffusion in the frequency axis direction in the time-frequency matrix due to the small fluctuations in rotation speed. Covering one-fifth of the frequency range on both sides of the rated frequency can completely include the main component of the target equipment's frequency and its frequency offset range caused by the rotation speed fluctuations. When the real-time rotation speed signal of the target equipment is available, the frequency neighborhood uses the real-time frequency corresponding to the real-time rotation speed instead of the rated frequency as the center frequency.

[0143] S5.2.2: Taking the integer multiple frequency position of the rated frequency as the center, determine the harmonic neighborhood corresponding to the second to fourth harmonics in sequence. The frequency width of each harmonic neighborhood is consistent with the width of the frequency range ...

[0144] S5.2.3: Take half the width of the rated frequency on both sides of the frequency range corresponding to the frequency range of the frequency conversion and harmonic neighborhood set as the sideband search range. The frequency range remaining after removing the frequency units covered by the frequency conversion and harmonic neighborhood set within the sideband search range is denoted as the sideband neighborhood. The sideband neighborhood is used to search for the modulation sideband components caused by local degradation of the target device.

[0145] It should be noted that the sideband neighborhood does not overlap with the set of frequency conversion and harmonic neighborhood. The frequency components in the sideband neighborhood reflect the amplitude modulation or frequency modulation effect of the modulation source inside the target device (such as loose bearing cage or tooth surface wear) on the main frequency conversion component. When the target device is in a healthy state, the energy in the sideband neighborhood is extremely low. When the target device enters a sub-healthy state, observable modulation sideband peaks appear in the sideband neighborhood.

[0146] S5.2.4: In the target source purification time-frequency matrix, the amplitude of each time frame is taken for the frequency row corresponding to the sideband neighborhood according to the time window. For each time frame, the energy ridge position is calculated in the sideband neighborhood in an amplitude-weighted manner. The energy ridge position is the weighted average of the center frequency of each frequency unit in the sideband neighborhood with the corresponding amplitude as the weight. The average of the energy ridge positions of each time frame in the current monitoring window is taken to obtain the sideband energy centroid frequency of the current monitoring window.

[0147] Specifically, the energy ridge position is calculated only for frequency cells whose amplitude exceeds the average amplitude of all frequency cells within the sideband neighborhood (i.e., all frequency cells in the frequency row covered by the sideband neighborhood in this time frame). Frequency cells with amplitudes lower than this average amplitude are not included in the weighted calculation to eliminate the interference of background noise components on the calculation results of the sideband energy centroid frequency in the current monitoring window. It is important to emphasize that the calculation range of the average amplitude is limited to the sideband neighborhood and does not include frequency cells in the frequency transition and harmonic neighborhood sets, nor does it include other frequency ranges outside the sideband neighborhood.

[0148] S5.2.5: Calculate the difference between the current monitoring window sideband energy centroid frequency and the switching sideband baseline value, and take the absolute value of the difference as the switching sideband drift.

[0149] In an optional embodiment, if the target device has a rated frequency of 50Hz, a sideband neighborhood covering 30Hz to 45Hz and 55Hz to 70Hz, a current monitoring window sideband energy centroid frequency of 58Hz, and a frequency conversion sideband baseline value of 51.2Hz, then the frequency conversion sideband drift is |58-51.2|=6.8Hz.

[0150] S5.3: Select a high-frequency impact band in the target source purification time-frequency matrix, and calculate the difference between the instantaneous energy peak value and the impact energy baseline value within the high-frequency impact band to obtain the residual impact energy.

[0151] S5.3.1: Based on the bearing structural parameters and rated rotational frequency of the target equipment, calculate the rolling element passing frequency, inner ring fault characteristic frequency and outer ring fault characteristic frequency of the bearing of the target equipment. The frequency range corresponding to half of the minimum value to twice the maximum value of the three is recorded as the high-frequency impact band.

[0152] Specifically, the bearing structural parameters include the bearing pitch diameter, rolling element diameter, number of rolling elements, and contact angle, which are read from the target equipment ledger.

[0153] It should be noted that the setting of the high-frequency impact band is based on the characteristic frequency of bearing failure, so that the high-frequency impact band covers the main frequency range of the periodic impact response caused by early pitting corrosion of the bearing. At the same time, the lower limit of the high-frequency impact band is higher than the highest frequency unit of the rotational frequency and harmonic neighborhood set, so as to avoid the energy of the main component of the rotational frequency being mixed into the statistical results of the high-frequency impact band. When the structural parameters of the bearing of the target equipment are not available, the high-frequency impact band is taken as the frequency range corresponding to five to twenty times the rated rotational frequency.

[0154] S5.3.2: Extract the frequency row corresponding to the high-frequency impact band from the target source purification time-frequency matrix, and sum the squares of the amplitudes of all frequency units in the high-frequency impact band for each time frame to obtain the high-frequency impact instantaneous energy sequence corresponding to each time frame.

[0155] S5.3.3: Take the maximum value of the high-frequency impact instantaneous energy sequence and record it as the high-frequency impact peak energy. Calculate the difference between the high-frequency impact peak energy and the impact energy baseline value in the historical health baseline parameter set. Record the part of the difference that is greater than zero as the impact energy residue. When the high-frequency impact peak energy is less than or equal to the impact energy baseline value, the impact energy residue is zero.

[0156] It should be noted that the portion of the difference in the residual impact energy is not less than zero because after the target source purification time-frequency matrix undergoes amplitude attenuation processing by S4, the energy in the high-frequency impact band is reduced compared to the original time-frequency matrix. When the reduced high-frequency impact peak energy is still higher than the impact energy baseline value, it indicates that the high-frequency impact component generated by the target equipment itself has exceeded the health benchmark level, and this excess is the residual impact energy. When the high-frequency impact peak energy is lower than the impact energy baseline value, it is determined that there is no observable impact anomaly in the current monitoring window, and the residual impact energy is taken as zero.

[0157] For example, if the peak energy of the high-frequency impact in the current monitoring window is 1.35 × 10⁻⁶, then... -4 m 2 / s 4 The baseline value for impact energy is 0.98 × 10⁻⁶. -4 m 2 / s 4 The residual impact energy is (1.35-0.98)×10 -4 =0.37×10 -4 m 2 / s 4 If the peak energy of the high-frequency impact in the current monitoring window is 0.85 × 10⁻⁶ -4 m 2 / s4 If the energy level is lower than the baseline value, the residual impact energy is taken as zero. S5.4: Select a broadband noise frequency band in the target source purification time-frequency matrix, calculate the average energy within the broadband noise frequency band, and compare the average energy with the broadband noise baseline value in the historical health baseline parameter set to obtain the broadband noise rise.

[0158] It should be noted that the broadband noise band is the continuous frequency range after removing the frequency transition neighborhood, sideband neighborhood, and high-frequency impulse band. Broadband noise rise is defined as: if the current average energy is greater than the broadband noise baseline value, the rise is expressed in decibels as the ratio of the current average energy to the baseline value. (Current average value / baseline value); if the current average energy is less than or equal to the broadband noise baseline value, the rise is zero.

[0159] S5.5: Arrange the frequency switching sideband drift, residual impact energy, and broadband noise rise in sequence to construct the original feature vector, and normalize it according to the standard deviation of the historical healthy baseline parameter set to obtain the sub-health coherent feature vector.

[0160] S5.5.1: Read the standard deviation of the frequency sideband baseline, the standard deviation of the impact energy baseline, and the standard deviation of the broadband noise baseline from the historical health baseline parameter set. These three parameters correspond to the statistical fluctuation range of the frequency sideband drift, the residual impact energy, and the broadband noise rise of the target equipment during the healthy operation phase, respectively.

[0161] It should be noted that the standard deviation of the frequency switching sideband baseline, the standard deviation of the impulse energy baseline, and the standard deviation of the broadband noise baseline are calculated simultaneously when establishing a historical health baseline parameter set during the healthy operation phase of the target equipment. The calculation is based on the sample standard deviation of the above three characteristics within no less than thirty monitoring windows. When the calculated result of a certain baseline standard deviation is zero or close to zero, the standard deviation of that baseline is taken as five percent of the corresponding baseline value to avoid division by zero during normalization.

[0162] S5.5.2: Divide the frequency switching sideband drift by the standard deviation of the frequency switching sideband baseline to obtain the normalized frequency switching sideband drift; divide the residual impulse energy by the standard deviation of the impulse energy baseline to obtain the normalized residual impulse energy; divide the broadband noise rise by the standard deviation of the broadband noise baseline to obtain the normalized broadband noise rise.

[0163] Specifically, the dimensions of the normalized frequency shift sideband drift, normalized residual impact energy, and normalized broadband noise rise are unified to multiples of their respective health baseline fluctuation ranges. A value of one indicates that the current characteristic deviation is equal to one standard deviation of the healthy operating stage, while a value of zero indicates that the current characteristic is consistent with the health baseline. The above normalization process eliminates the dimensional differences between the frequency shift sideband drift (measured in Hertz), the residual impact energy (measured in Energy), and the broadband noise rise (measured in Decibels), ensuring that the three have equal weight in the subsequent deviation calculation.

[0164] S5.5.3: Normalize the transition sideband drift. Normalized impact energy residue and normalized broadband noise rise Arranged in order, they form a three-dimensional real-valued vector, denoted as the sub-health-related feature vector. .

[0165] It should be noted that the three components of the sub-health coherent feature vector correspond to the degree of feature deviation under the three sub-health degradation modes of the target device: modulation-type degradation, impact-type degradation, and distributed degradation. The three are independent of each other in terms of physical degradation mechanism; the sub-health coherent feature vector in the healthy state should be a zero vector.

[0166] In an optional embodiment, if the current monitoring window's frequency shift sideband drift is 6.8 Hz and the frequency shift sideband baseline standard deviation is 1.2 Hz, then the normalized frequency shift sideband drift is 6.8 ÷ 1.2 ≈ 5.67; the residual impact energy is 0.37 × 10⁻⁶. -4 m 2 / s 4 The baseline standard deviation of the impact energy is 0.15 × 10⁻⁶. -4 m 2 / s 4 The normalized residual impact energy is 0.37÷0.15≈2.47; the broadband noise rise is 1.8dB and the broadband noise baseline standard deviation is 0.6dB, so the normalized broadband noise rise is 1.8÷0.6=3.0; the sub-health coherent feature vector is [5.67,2.47,3.0].

[0167] S5.6: Calculate the Euclidean norm of the sub-health-related feature vector, denoted as the comprehensive deviation of this monitoring window.

[0168] It should be noted that the Euclidean norm represents the overall deviation of the target equipment from the healthy operation stage within the current monitoring window; the sub-healthy coherent feature vector in the healthy state is a zero vector with a norm of zero; as the equipment degrades, each component increases, and the norm increases accordingly.

[0169] S5.7: Calculate the comprehensive deviation within the continuous monitoring window according to the time sequence, and compare the comprehensive deviation with the preset deviation threshold to determine whether the target equipment is in a sub-healthy state.

[0170] S5.7.1: When the comprehensive deviation within a consecutive preset number of monitoring windows is greater than the preset deviation threshold, the target device is determined to be in a sub-healthy state, and the target device number, sub-healthy coherent feature vector, and corresponding monitoring window are output.

[0171] S5.7.1.1: Read the preset deviation threshold and preset quantity.

[0172] It should be noted that the preset deviation threshold is obtained by multiplying the maximum value of the comprehensive deviation of the target equipment in each monitoring window during the healthy operation phase by a preset margin coefficient; the preset margin coefficient ranges from 1.2 to 1.5; the preset number is three to five monitoring windows; the purpose of setting the preset deviation threshold and the preset margin coefficient is to reserve a margin for the normal fluctuation of the comprehensive deviation during the healthy operation phase, and to prevent a single occasional vibration disturbance from triggering the sub-health state judgment; the purpose of setting the preset number to three to five monitoring windows is to require the comprehensive deviation to continuously exceed the preset deviation threshold in multiple consecutive monitoring windows, so as to eliminate misjudgments caused by occasional interference; the preset deviation threshold is divided into a first-level deviation threshold and a second-level deviation threshold, and the second-level deviation threshold is 1.5 to 2 times the first-level deviation threshold.

[0173] Specifically, when the overall deviation exceeds the first-level deviation threshold, a mild sub-health condition is determined; when the overall deviation exceeds the second-level deviation threshold, a severe sub-health condition is determined.

[0174] S5.7.1.2: Establish a continuous monitoring window counter and compare the comprehensive deviation calculated for each new monitoring window with the preset deviation threshold. If the comprehensive deviation is greater than the preset deviation threshold, increment the value of the continuous monitoring window counter by one; if the comprehensive deviation is less than or equal to the preset deviation threshold, reset the value of the continuous monitoring window counter to zero.

[0175] It should be noted that the initial value of the continuous monitoring window counter is zero; the continuous monitoring window counter records the number of consecutive over-threshold monitoring windows traced back to the current moment. The reset of the counter value ensures that non-continuous and scattered over-threshold windows do not accumulate and trigger the sub-health status judgment; the continuous monitoring window counter and the sub-health coherent feature vector are stored according to the monitoring window sequence number, so as to trace the feature data of the specific monitoring window in subsequent output.

[0176] S5.7.1.3: When the value of the continuous monitoring window counter reaches the preset number, extract the sub-health coherent feature vectors corresponding to the current monitoring window and the previous continuous over-threshold monitoring windows, read the target device number and the corresponding monitoring window number, and output the target device number, the sub-health coherent feature vectors corresponding to each continuous over-threshold monitoring window and the monitoring window number as the sub-health status determination result.

[0177] Specifically, the output of the sub-health status determination result includes: target device number, sub-health status level, continuous over-threshold starting monitoring window sequence number, current monitoring window sequence number, three component values ​​of the current sub-health coherent feature vector, and comprehensive deviation value; the three component values ​​of the sub-health coherent feature vector include normalized frequency switching sideband drift, normalized impact energy residue, and normalized broadband noise rise; each component value in the sub-health status determination result is used to assist in determining the dominant degradation mode of the target device. If the normalized frequency switching sideband drift is the largest among the three component values ​​of the sub-health coherent feature vector, the target device is determined to be dominated by modulation-type degradation; if the normalized impact energy residue is the largest among the three component values ​​of the sub-health coherent feature vector, the target device is determined to be dominated by impact-type degradation; if the normalized broadband noise rise is the largest among the three component values ​​of the sub-health coherent feature vector, the target device is determined to be dominated by distributed degradation.

[0178] In an optional embodiment, if the preset number is three monitoring windows, the comprehensive deviations corresponding to the seventh, eighth, and ninth monitoring windows are 4.21, 4.53, and 4.78, respectively, and the preset deviation threshold is 3.50, then the continuous monitoring window counter reaches three in the ninth monitoring window, triggering the sub-health state determination, and outputting the target device number, the sub-health state level as mild sub-health, the starting monitoring window number as seven, the current monitoring window number as nine, the current sub-health coherent feature vector as [5.67, 2.47, 3.0], and the comprehensive deviation as 4.78.

[0179] S5.7.2: When there is at least one monitoring window within a continuous preset number of monitoring windows whose comprehensive deviation is less than or equal to the preset deviation threshold, this monitoring window is recorded as a normal window, and the continuous monitoring window counter is reset to zero and the continuous monitoring window counter is re-accumulated.

[0180] It should be noted that the purpose of storing the sub-health coherent feature vector and comprehensive deviation of the normal window into the historical deviation record is to provide data support for the rolling update of the preset deviation threshold. After the continuous monitoring window counter is reset to zero, the subsequent monitoring windows need to start from zero again to continuously accumulate the over-threshold count to ensure the continuity requirement of the sub-health status judgment.

[0181] In summary, this invention clarifies the source and propagation path of vibration interference from adjacent equipment by synchronously acquiring vibration signals from the target equipment and adjacent equipment and recording the length of the support connection path. By calculating the structural propagation delay time and generating a structural propagation delay sequence, the vibration signals of adjacent equipment can be accurately aligned with the vibration signals of the target equipment in the time-frequency domain. A time-frequency matrix is ​​constructed through short-time Fourier transform, and the propagation interference region is identified based on the local coherence coefficient. The amplitude of the interference components in the time-frequency matrix of the target equipment is attenuated to obtain a target source purification time-frequency matrix that better reflects the state of the target equipment itself. Furthermore, the frequency shift, residual impact energy, and broadband noise rise are extracted from the target source purification time-frequency matrix to construct a sub-health coherent feature vector. This vector is then combined with the comprehensive deviation of the continuous monitoring window for judgment, thereby reducing misjudgments caused by coupled vibrations of adjacent equipment and improving the accuracy, stability, and interpretability of identifying the sub-health state of industrial pipe gallery equipment.

[0182] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for identifying the sub-health state of industrial pipe gallery equipment based on time-frequency domain fusion features, characterized in that, include: Accelerometers are installed at the bearing housing of the target equipment and at the connection points of the brackets from each adjacent equipment to the target equipment. Vibration signals of the target equipment and adjacent equipment are collected synchronously, and the set of bracket connection path lengths is recorded. Based on the set of support connection path lengths and the longitudinal wave velocity of the support material, the structural propagation delay time of vibration signals from each adjacent device to the target device measurement point is calculated, and a structural propagation delay sequence is generated. Short-time Fourier transforms are performed on the vibration signals of the target device and the adjacent device respectively to obtain the time-frequency matrix of the target device and the time-frequency matrix of the adjacent device, and the time-frequency matrix of the adjacent device is delayed and aligned according to the structure propagation delay sequence. Calculate the local coherence coefficients of the target device time-frequency matrix and the time-frequency matrices of adjacent devices after delay alignment, and perform amplitude attenuation on the propagation interference region in the target device time-frequency matrix according to the local coherence coefficients to obtain the target source purification time-frequency matrix; Extract the frequency shift, residual impact energy, and broadband noise rise from the target source purification time-frequency matrix to construct a sub-health coherent feature vector and determine the sub-health state of the target equipment.

2. The method for identifying sub-health status of industrial pipe gallery equipment based on time-frequency domain fusion features as described in claim 1, characterized in that, Constructing a sub-health coherent feature vector to determine the sub-health state of the target device includes: Read the target source purification time-frequency matrix and call the target device's rated frequency, historical health baseline parameter set, and preset deviation threshold; Centered on the rated frequency, select the frequency-revolution neighborhood and the sideband neighborhood in the target source purification time-frequency matrix, extract the energy ridge line position in the frequency-revolution neighborhood and the sideband neighborhood, and take the offset of the energy ridge line position relative to the frequency-revolution sideband baseline value in the current monitoring window as the frequency-revolution sideband drift amount; In the target source purification time-frequency matrix, a high-frequency impact band is selected, and the difference between the instantaneous energy peak value and the impact energy baseline value within the high-frequency impact band is calculated to obtain the residual impact energy. A broadband noise frequency band is selected in the target source purification time-frequency matrix, the average energy in the broadband noise frequency band is calculated, and the average energy is compared with the broadband noise baseline value in the historical health baseline parameter set to obtain the broadband noise rise amount. The frequency switching sideband drift, the residual impact energy, and the broadband noise rise are arranged in order to construct the original feature vector, and then normalized according to the standard deviation of the historical health baseline parameter set to obtain the sub-health coherent feature vector.

3. The method for identifying sub-health status of industrial pipe gallery equipment based on time-frequency domain fusion features as described in claim 2, characterized in that, It also includes, Calculate the Euclidean norm of the sub-health coherent feature vector, and denote it as the comprehensive deviation of this monitoring window; The comprehensive deviation within the continuous monitoring window is statistically analyzed in chronological order, and the comprehensive deviation is compared with a preset deviation threshold to determine whether the target device is in a sub-healthy state. When the comprehensive deviation amount within a consecutive preset number of monitoring windows is greater than the preset deviation threshold, the target device is determined to be in a sub-healthy state, and the target device number, the sub-healthy coherent feature vector, and the corresponding monitoring window are output. When the comprehensive deviation of at least one monitoring window within a consecutive preset number of monitoring windows is less than or equal to the preset deviation threshold, this monitoring window is recorded as a normal window, and the continuous monitoring window counter is reset to zero and the continuous monitoring window counter is re-accumulated.

4. The method for identifying sub-health status of industrial pipe gallery equipment based on time-frequency domain fusion features as described in claim 3, characterized in that, In the sub-health state determination result, each component value is used to assist in determining the dominant degradation mode of the target device. If the normalized switching sideband drift is the largest among the three component values ​​of the sub-health coherent feature vector, the target device is determined to be dominated by modulation-type degradation. If the normalized residual impulse energy is the largest among the three component values ​​of the sub-health coherent feature vector, the target device is determined to be dominated by impulse-type degradation. If the normalized broadband noise rise is the largest among the three component values ​​of the sub-health coherent feature vector, the target device is determined to be dominated by distributed degradation. The three component values ​​of the sub-health coherent feature vector include the normalized switching sideband drift, the normalized residual impulse energy, and the normalized broadband noise rise.

5. The method for identifying sub-health status of industrial pipe gallery equipment based on time-frequency domain fusion features as described in claim 2, characterized in that, The target source purification time-frequency matrix includes: The propagation interference determination coefficient is compared with a preset coherence threshold. If the propagation interference determination coefficient is greater than the preset coherence threshold, the frequency unit is marked as a propagation interference unit, and all the propagation interference units constitute a propagation interference region. If the propagation interference determination coefficient is less than or equal to the preset coherence threshold, the frequency unit is marked as a non-propagation interference unit. The amplitude attenuation coefficient is calculated based on the propagation interference determination coefficient, and the amplitude of the propagation interference region in the time-frequency matrix of the target device is attenuated according to the amplitude attenuation coefficient. The amplitude and phase of the non-propagation interference region in the time-frequency matrix of the target device are retained, and the propagation interference region after amplitude attenuation is merged with the non-propagation interference region to obtain the target source purification time-frequency matrix.

6. The method for identifying sub-health status of industrial pipe gallery equipment based on time-frequency domain fusion features as described in claim 5, characterized in that, The propagation interference determination coefficient and the local coherence coefficient include: Read the target device time-frequency matrix and the time-frequency matrix aligned with each adjacent device, and confirm that the target device time-frequency matrix and the time-frequency matrix aligned with each adjacent device have the same frequency unit and time window; Taking each time-frequency unit in the target device's time-frequency matrix as the center, a preset number of adjacent frequency units and adjacent time windows are selected to construct a local time-frequency neighborhood; Within each of the local time-frequency neighborhoods, the cross power spectrum and self power spectrum between the time-frequency matrix of the target device and the aligned time-frequency matrix of the adjacent device are calculated respectively. Based on the cross power spectrum and the self power spectrum, calculate the local coherence coefficients of the target device time-frequency matrix and the time-frequency matrix aligned with each of the adjacent devices in each of the local time-frequency neighborhoods; For the same time-frequency unit, the local coherence coefficients corresponding to the aligned time-frequency matrices of each adjacent device are compared, and the largest local coherence coefficient is selected as the propagation interference determination coefficient of that time-frequency unit.

7. The method for identifying sub-health status of industrial pipe gallery equipment based on time-frequency domain fusion features as described in claim 6, characterized in that, The target device time-frequency matrix and the adjacent device time-frequency matrix include: Read the vibration signal of the target device, the vibration signal of the adjacent device, and the structural propagation delay sequence, wherein the structural propagation delay sequence includes the structural propagation delay time and the number of sampling point delays corresponding to each adjacent device number; Configure the same short-time Fourier transform parameter set for the vibration signal of the target device and the vibration signal of adjacent devices; According to the short-time Fourier transform parameter set, the vibration signal of the target device is subjected to frame windowing and short-time Fourier transform to obtain the time-frequency matrix of the target device, wherein the rows of the time-frequency matrix of the target device correspond to frequency units and the columns correspond to time windows.

8. The method for identifying sub-health status of industrial pipe gallery equipment based on time-frequency domain fusion features as described in claim 7, characterized in that, According to the short-time Fourier transform parameter set, the vibration signals of each adjacent device are subjected to frame windowing and short-time Fourier transform respectively to obtain the time-frequency matrix of each adjacent device.

9. The method for identifying sub-health status of industrial pipe gallery equipment based on time-frequency domain fusion features as described in claim 8, characterized in that, The structure propagation delay sequence includes: Read the set of support connection path lengths and retrieve the longitudinal wave velocity of the support material corresponding to the installation section of the target equipment; According to the adjacent device number, read the corresponding bracket connection path length item by item from the bracket connection path length set; The structural propagation delay time of the vibration signal of each adjacent device to the measuring point of the target device is obtained by calculating the ratio of the length of the connection path of each support to the longitudinal wave velocity of the support material. Based on the sampling frequencies of the vibration signals of the target device and the adjacent devices, the propagation delay time of each structure is converted into the corresponding sampling point delay number; Establish a correspondence between the adjacent device numbers, the structure propagation delay time, and the number of sampling point delays to form a structure propagation delay record; Sort all the structure propagation delay records according to the order of the adjacent device numbers to generate a structure propagation delay sequence.

10. The method for identifying sub-health status of industrial pipe gallery equipment based on time-frequency domain fusion features as described in claim 9, characterized in that, The length of the support connection path is the actual path length of the vibration signal of the adjacent equipment transmitted to the measuring point of the target equipment along the continuous connection direction of the support.

Images