Power plant induced draft fan fault early warning method and system based on current-voiceprint

Abstract

The invention discloses a power plant induced draft fan fault early warning method and system based on current-voiceprint. The method comprises the steps that first data and second data of an induced draft fan are collected; carrying out autocorrelation matrix and cross-correlation vector calculation on the two groups of preprocessed data, and solving a tap weight vector of a first filter; based on the tap weight vector, solving a mean square error of a real-time effective value and an estimated value of the second data, and determining a first filter order; combining the first filter order with the weight vector, calculating a deviation degree and converting the deviation degree into component values; and generating an equipment health degree value based on the score in combination with a scoring mechanism. According to the invention, the maintenance cost is reduced, forced load drop and even stop of the generator set caused by forced failure of the equipment are avoided, and the economic benefit and the reliability of the generator set are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault warning, and in particular to a method and system for warning faults of induced draft fans in power plants based on current-voiceprint. Background Art

[0002] One of the important tasks of daily inspection of induced draft fans in power plants is for staff to take a listening stick to key locations on site to judge the health status of the induced draft fans by listening. There are several problems here:

[0003] 1. It is impossible to achieve 24-hour uninterrupted monitoring by manual methods;

[0004] 2. Inspection quality depends entirely on the skill level and experience of the inspector, which is an uncontrollable element;

[0005] 3. The same induced draft fan with the same health level will produce different soundprint signals when the induced draft fan is in a light load or heavy load state, or when the induced draft fan is increasing or decreasing its load, and the manual method cannot synchronously grasp the corresponding state of the induced draft fan.

[0006] 4. Manual inspection methods can only rely on experience and feelings, and cannot be described quantitatively; two people with the same experience and skills cannot communicate through quantitative description methods. In other words, manual methods cannot provide a detailed description of the equipment status. Summary of the invention

[0007] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0008] In view of the above existing problems, the present invention is proposed.

[0009] Therefore, the present invention provides a method and system for early warning of induced draft fan faults in power plants based on current-voiceprint to solve the problem that the existing early warning methods are prone to misjudgment or omission of faults.

[0010] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0011] In a first aspect, the present invention provides a method for early warning of induced draft fan failure in a power plant based on current-voiceprint, comprising:

[0012] Collecting first data and second data of the induced draft fan;

[0013] The autocorrelation matrix and the cross-correlation vector are calculated for the two sets of preprocessed data to solve the tap weight vector of the first filter;

[0014] Based on the tap weight vector, solving the mean square error between the real-time effective value of the second data and the estimated value, and determining the order of the first filter;

[0015] The first filter order is combined with the weight vector, the deviation is calculated and converted into a component value;

[0016] Based on the score and combined with the scoring mechanism, a device health value is generated.

[0017] As a preferred solution of the power plant induced draft fan fault early warning method based on current-soundprint described in the present invention, wherein:

[0018] The preprocessing comprises:

[0019] Perform anti-aliasing filtering on the data of all channels;

[0020] Assume that the first channel collects the first data, and the second channel collects the second data;

[0021] The first data is a current signal of the induced draft fan driving motor, and the second data is a sound print signal of a key position;

[0022] Based on the current signal and the voiceprint signal after anti-aliasing filtering, the effective value of each channel signal is calculated to obtain a first data effective value sequence and a second data effective value sequence.

[0023] As a preferred solution of the power plant induced draft fan fault early warning method based on current-soundprint described in the present invention, wherein:

[0024] The second-order statistics estimation comprises:

[0025] Calculate the M-order autocorrelation matrix of the first data effective value sequence;

[0026] An M-th order cross-correlation vector between the first data and the second data is calculated.

[0027] As a preferred solution of the power plant induced draft fan fault early warning method based on current-soundprint described in the present invention, wherein:

[0028] The step of calculating the M-order autocorrelation matrix of the first data effective value sequence includes:

[0029] For a given positive integer order M, calculate the autocorrelation function;

[0030] Based on the results of the autocorrelation function, an M-order autocorrelation matrix is ​​constructed;

[0031] The calculating the M-order cross-correlation vector between the first data and the second data comprises:

[0032] For a given positive integer order M, calculate the cross-correlation function;

[0033] According to the result of the cross-correlation function, an M-order cross-correlation function sequence is obtained;

[0034] The cross-correlation function sequence is combined into a column vector, namely, the M-order cross-correlation vector.

[0035] As a preferred solution of the power plant induced draft fan fault early warning method based on current-soundprint described in the present invention, wherein:

[0036] The step of solving the tap weight vector of the first filter comprises:

[0037] Based on the M-order autocorrelation matrix of the first data effective value sequence, calculating the inverse matrix of the autocorrelation matrix;

[0038] The tap weight vector is calculated by applying the inverse matrix of the autocorrelation matrix to the M-order cross-correlation vector;

[0039] Based on the tap weight vector, the tap weight vector that minimizes the mean square error is selected as the optimal tap weight vector.

[0040] As a preferred solution of the power plant induced draft fan fault early warning method based on current-soundprint described in the present invention, wherein:

[0041] The step of determining the first filter order by solving the mean square error between the real-time effective value of the second data and the estimated value comprises:

[0042] Calculate an estimated value of the real-time effective value of the second data by transposing the optimal tap weight vector;

[0043] Calculate the mean square error between the real-time effective value of the second data and the estimated value;

[0044] It is determined whether the mean square error is less than a first threshold value. If so, the mean square error is the optimal order of the first filter.

[0045] As a preferred solution of the power plant induced draft fan fault early warning method based on current-soundprint described in the present invention, wherein:

[0046] The step of combining the first filter order with the weight vector, calculating the deviation and converting it into a component value comprises the following steps:

[0047] Based on the optimal order of the first filter and in combination with the optimal tap weight vector, mapping is performed to obtain the average model point within a preset time period;

[0048] Based on the optimal order of the first filter and in combination with the optimal weight vector, the dynamic model sequence points within the preset time step are mapped;

[0049] For each dynamic model point in the dynamic model sequence points, the Euclidean distance between each dynamic model point and the average model point is calculated, and the maximum Euclidean distance from all dynamic model points to the average model point is determined;

[0050] For the new real-time monitoring data collected within the preset time step, the corresponding optimal tap weight vector is calculated and mapped to obtain the real-time monitoring data mapping points;

[0051] Compare the real-time monitoring data mapping point with the average model point to obtain the Euclidean distance between the real-time monitoring data mapping point and the average model point;

[0052] Based on the maximum Euclidean distance from all dynamic model points to the average model point and the Euclidean distance between the real-time monitoring data mapping point and the average model point, it is converted into a score according to the scoring formula.

[0053] In a second aspect, the present invention provides a power plant induced draft fan fault early warning system based on current-voiceprint, comprising:

[0054] A data acquisition module, used for acquiring first data and second data of the induced draft fan;

[0055] A solution module, used for calculating the autocorrelation matrix and the cross-correlation vector of the two sets of preprocessed data, and solving the tap weight vector of the first filter;

[0056] A filter order determination module, used to solve the mean square error between the real-time effective value and the estimated value of the second data based on the tap weight vector, and determine the first filter order;

[0057] A calculation module, used for combining the first filter order with the weight vector, calculating the deviation and converting it into a component value;

[0058] The generation module is used to generate a device health value based on the score and in combination with the scoring mechanism.

[0059] In a third aspect, the present invention provides a computing device, comprising:

[0060] Memory, used to store programs;

[0061] A processor is used to execute the computer executable instructions, which, when executed by the processor, implement the steps of the current-voiceprint based power plant induced draft fan fault early warning method.

[0062] In a fourth aspect, the present invention provides a computer-readable storage medium, comprising: when the program is executed by a processor, the steps of implementing the current-voiceprint based power plant induced draft fan fault warning method.

[0063] Beneficial effects of the present invention:

[0064] 1. By monitoring the real-time online voiceprint signal of the equipment for abnormal warning, it replaces the traditional method of manual on-site monitoring with ears through listening sticks, saving labor costs compared with the traditional method.

[0065] 2. The synchronous collection and fusion modeling of voiceprint signals and device drive motor current signals can determine whether the voiceprint signals are abnormal under different device load conditions, which is more accurate than the determination of pure voiceprint signals.

[0066] 3. Use a digital filter to convert the current of a real device driving the motor into a voiceprint signal. This effect is further mathematically described by a filter parameter set, so that the filter parameter set is used as a digital model of the device.

[0067] 4. The digital model obtained by the data of continuous operation for 72 hours after the equipment overhaul is used as the digital model of the initial good state of the equipment. In the process of monitoring operation, the data of each hour is modeled to obtain the real-time dynamic digital model of the equipment. The difference between the real-time dynamic model and the digital model of the good state describes the degree of deterioration of the health state of the equipment, and is quantitatively described by linearly transforming the distance in the M space to a score. This is consistent with the traditional maintenance experience that takes the equipment just completed as the starting point and then the equipment gradually ages, and also provides a score to describe the degree of deterioration of the health state of the equipment, which is conducive to the maintenance personnel to truly grasp the degree of urgency and arrange maintenance in a planned manner according to objective conditions (whether the equipment can be decommissioned, whether there are spare parts), so as to avoid "fire-fighting repair" due to sudden equipment failure and "over-repair" when the equipment is found to be intact after disassembly. It is conducive to reducing maintenance costs, avoiding the forced load reduction or even shutdown of the generator set caused by the forced failure of the equipment, and improving the economic benefits and reliability of the generator set. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0069] Figure 1 A basic flow chart of a method for early warning of induced draft fan failure in a power plant based on current-voiceprint provided by an embodiment of the present invention;

[0070] Figure 2 A device input-output relationship diagram of a power plant induced draft fan fault early warning method based on current-voiceprint provided by an embodiment of the present invention;

[0071] Figure 3A mathematical model error evaluation block diagram of a method for early warning of induced draft fan failure in a power plant based on current-voiceprint provided by an embodiment of the present invention;

[0072] Figure 4 A flow chart of determining the filter order of a method for early warning of a power plant induced draft fan fault based on current-voiceprint provided by one embodiment of the present invention;

[0073] Figure 5 A curve diagram of mean square error J(M) of a power plant induced draft fan fault early warning method based on current-voiceprint provided by one embodiment of the present invention with respect to order M;

[0074] Figure 6 A flow chart of establishing an equipment model for a method for early warning of a power plant induced draft fan fault based on current-voiceprint provided by an embodiment of the present invention;

[0075] Figure 7 A real-time monitoring flow chart of a method for early warning of a power plant induced draft fan failure based on current-voiceprint provided by an embodiment of the present invention;

[0076] Figure 8 A real-time curve graph of scores of a method for early warning of induced draft fan failure in a power plant based on current-voiceprint provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0077] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0078] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0079] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0080] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0081] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0082] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0083] Example 1

[0084] Reference Figure 1 , which is an embodiment of the present invention, provides a method for early warning of induced draft fan failure in a power plant based on current-voiceprint, comprising:

[0085] S1: collecting first data and second data of the induced draft fan;

[0086] S2: Calculate the autocorrelation matrix and cross-correlation vector of the two sets of preprocessed data to solve the tap weight vector of the first filter;

[0087] S3: Based on the tap weight vector, solving the mean square error between the real-time effective value of the second data and the estimated value, and determining the order of the first filter;

[0088] S4: Combine the first filter order with the weight vector, calculate the deviation and convert it into a component value;

[0089] S5: Generate a device health value based on the score and combined with the scoring mechanism.

[0090] It should be noted that the current signal of the induced draft fan driving motor is taken as input, and the signal of the soundprint sensor measurement point is taken as the expected response. The entire induced draft fan electromechanical system is regarded as an FIR filter, and the tap weight coefficient group of the filter is the FIR filter design, thereby realizing the mapping of the real device to the FIR filter, and the order of the FIR filter and the tap weight coefficient group constitute the digital modeling of the real device. The error between the digital model and the real device is characterized by the minimum mean square error. Correspondingly, the process of the equipment slowly aging from completely healthy to finally failing and unable to continue to be used, the corresponding digital model constantly deviates from the digital model when it is completely healthy. The digital model corresponding to the point in the multidimensional space and the distance between the points are used to describe the difference measure between the two different digital models, so as to give the difference change curve of the equipment from completely healthy and slowly aging to failing and unable to continue to be used, and give the warning domain to remind the relevant maintenance and operation personnel.

[0091] Example 2

[0092] Reference Figure 2-8 , which is an embodiment of the present invention, provides a power plant induced draft fan fault early warning method based on current-voiceprint based on the previous embodiment, comprising:

[0093] In an optional embodiment, the first data and the second data in S1 may be a current signal and a voiceprint signal, or a vibration signal and a temperature signal, or a pressure signal and a flow signal;

[0094] In an optional embodiment, the current signal and the voiceprint signal include the current signal of the induced draft fan driving motor as input and the voiceprint sensor measurement point signal as the expected response. The entire induced draft fan's electromechanical system is regarded as a FIR filter, and the tap weight coefficient group of the filter is the FIR filter design, thereby realizing the mapping of the real device to the FIR filter, and the order of the FIR filter and the tap weight coefficient group constitute the digital modeling of the real device.

[0095] In an optional embodiment, the vibration signal and the temperature signal include taking the vibration signal as input and the temperature signal as the expected response, the vibration signal detecting wear, imbalance, looseness and other problems of mechanical parts, and the temperature signal monitoring the operating temperature of the equipment and identifying overheating problems.

[0096] In an optional embodiment, the pressure signal and flow signal include a pressure signal as an input and a flow signal as an expected response. The pressure signal monitors the pressure changes of the system to ensure that it operates within a safe range and prevents leakage or other failures. The flow signal ensures the stability of the system flow and identifies problems such as blockages or leakage.

[0097] It should be noted that the present invention selects current and voiceprint signals because they provide information in two dimensions, electrical and mechanical, which complement each other. For example, a current anomaly may indicate an electrical problem, while a voiceprint anomaly may indicate a mechanical problem. The combination of current and voiceprint signals can more accurately locate the cause of the fault and reduce false alarms. Although vibration signals can reflect mechanical problems, accelerometers need to be installed, which increases hardware costs and installation complexity. Although temperature signals are important, a single temperature change may not directly reflect the specific type of fault, and the sensitivity is not as good as current and voiceprint signals. Pressure and flow signals are suitable for fluid systems, but they are not direct enough for detecting electrical and mechanical faults in mechanical equipment. The installation and maintenance costs are high, and the data processing complexity is large.

[0098] In the embodiment of the present application, the multi-channel data acquisition card in S1 synchronously collects the current of the induced draft fan drive motor and the sound print sensor signal installed at the key position (can be multiple), the data sampling frequency is not less than 25000Hz, and the data acquisition method is continuous sampling and real-time transmission to the data server. Suppose the first channel is the current signal, and the second channel is the sound print signal of a key position. Here, the data collected after the equipment is overhauled and continuously running for 72 hours is generally used as the modeling data, and the data collected every hour thereafter is used as the dynamic model data.

[0099] In an embodiment of the present application, preprocessing includes anti-aliasing filtering of data from all channels; assuming that the first channel collects first data and the second channel collects second data; the first data is the current signal of the induced draft fan drive motor, and the second data is a soundprint signal at a key position; based on the current signal and the soundprint signal after anti-aliasing filtering, the effective value of each channel signal is calculated to obtain a first data effective value sequence and a second data effective value sequence.

[0100] In the embodiment of the present application, anti-aliasing filtering includes anti-aliasing filtering (low-pass filtering with a cutoff frequency of half the sampling frequency) on the data of all channels. Assume that after anti-aliasing filtering, the current signal obtained by the on-site acquisition device is a discrete sequence:

[0101] s 1 (0),s 1 (1),s 1 (2),s 1 (3),s 1 (4),…;

[0102] The obtained voiceprint signal is a discrete sequence:

[0103] s 2 (0),s 2 (1),s 2 (2),s 2 (3),s 2 (4),…;

[0104] In the embodiment of the present application, finding the effective value includes assuming that: the sampling frequency is f s ,

[0105] For s 1 (0),s 1 (1),s 1 (2),s 1 (3),s 1 (4),…,Let:

[0106]

[0107] The real-time effective value sequence of the current signal is obtained and recorded as:

[0108] S 1 =S 1 (0),S 1 (1),S 1 (2),S 1 (3),S 1 (4),…,S 1 (N-1)(2)

[0109] For s 2 (0),s 2 (1),s 2 (2),s 2 (3),s 2 (4),…,Let:

[0110]

[0111] The real-time effective value sequence of the voiceprint signal is obtained and recorded as:

[0112] S 2 =S 2 (0),S 2 (1),S 2 (2),S 2 (3),S 2 (4),…,S 2 (N-1)(4)

[0113] Among them, S represents the effective value, s represents the instantaneous value, and N represents the length of the sequence.

[0114] It should be noted that by synchronously collecting the real-time current signal of the same device and the voiceprint signal at key positions, and coupling the data of the two channels by establishing a cross-correlation matrix in the algorithm, the abnormal conditional judgment of the voiceprint signal of the device under different load conditions (driving motor current) is realized. It avoids the inability to take into account the current signal at the same time in manual and traditional methods.

[0115] In an optional embodiment, the first filter may be a FIR filter, an IIR filter, or an adaptive filter;

[0116] In an optional embodiment, the scheme of selecting the first filter as a FIR filter is to simplify the electromechanical system of the induced draft fan into a FIR filter model, and use the minimum mean square error (LMS) algorithm to adjust the tap weights of the FIR filter according to the collected current and voiceprint data, so that the model output matches the actual measurement value as much as possible, and by comparing the changes in FIR model parameters at different time points, the aging degree of the equipment is quantified, and a difference change curve from healthy to faulty is plotted, and a warning domain is set. When the model parameters deviate from the normal range, an early warning is automatically issued to alert maintenance personnel.

[0117] In an optional embodiment, the scheme of selecting the first filter as an IIR filter is to establish a preliminary IIR filter model based on the physical characteristics and expected working conditions of the induced draft fan electromechanical system, and use the minimum mean square error algorithm or the gradient descent algorithm to adjust the coefficient of the IIR filter according to the collected motor current signal and the soundprint sensor measurement point signal. By continuously monitoring the change trend of these parameters, the health status of the equipment can be evaluated. When the parameters deviate from the preset safety range, an alarm is triggered to prompt the maintenance personnel to check.

[0118] In an optional embodiment, the scheme for selecting the first filter as an adaptive filter is to determine which adaptive algorithm to use to update the weight of the filter. Common choices include but are not limited to LMS algorithm, recursive least squares (RLS), Kalman filtering, etc. Automatically adjust its own parameters based on real-time data. After deploying the adaptive filter, continuously receive new input data and continuously adjust its own parameters to minimize the prediction error. By comparing the difference between the output of the adaptive filter and the actual measurement, and observing the changes in the filter parameters over time, long-term tracking of the health status of the equipment is achieved. Once an abnormal pattern is found or a situation that exceeds the set threshold is detected, a warning is issued to notify relevant personnel.

[0119] It should be noted that the reason why we chose FIR filter in our invention is that it is unconditionally stable, depends only on the input signal, and has no internal feedback loop. This inherent stability simplifies the design process and improves the reliability of the system. It can be easily designed to have a strict linear phase characteristic, which is essential for maintaining the time alignment of the signal, especially when processing the signal of the voiceprint sensor measurement point, to ensure that the original characteristics of the signal are not destroyed. The design is relatively simple and can be easily implemented through standard algorithms such as window method, frequency sampling method, etc., reducing the design difficulty and time cost. Because IIR filters contain feedback paths, they may become unstable, especially in the case of improper design or unreasonable parameter selection. This increases the complexity and risk of system design. In order to adjust the filter coefficients in real time, adaptive filters often require higher computing power, which may lead to higher hardware costs and power consumption.

[0120] In the embodiment of the present application, the calculation of the autocorrelation matrix and the cross-correlation vector in S2 includes finding S 1 The M-order autocorrelation matrix And find the M-order cross-correlation vector of the first data and the second data

[0121] In the present embodiment, S 1 The M-order autocorrelation matrix For a given order M, M is a positive integer, define For S 1 The M-order autocorrelation matrix is ​​as follows:

[0122] In the embodiment of the present application, the autocorrelation function is calculated:

[0123]

[0124] In the formula, Indicates S 1 The M-order autocorrelation function, where the subscript 1,1 represents the autocorrelation between the first channel and the first channel, and the superscript (M) represents the filter order. The n in the brackets represents the fixed delay in the time domain, which can be 0, 1, 2, ..., M-1, a total of M, which is equal to the filter order; S 1 (m) represents S 1 The mth data; S 1 (m+n) represents S 1 The m+nth data; Σ represents the summation symbol, whose lower limit is 0 and upper limit is NM.

[0125] The M-order autocorrelation function sequence is obtained from formula (5):

[0126] A total of M.

[0127] In the embodiment of the present application, the M-order autocorrelation matrix is ​​calculated:

[0128]

[0129] In the embodiment of the present application, the M-order cross-correlation vector of the first data and the second data is calculated. include:

[0130] For S defined in formula (2) 1 and S defined by formula (4) 2 , calculate the cross-correlation function:

[0131]

[0132] In the formula, The autocorrelation function of the real-time effective value of the first channel and the second channel is represented, where the subscripts 1 and 2 represent the cross-correlation between the first channel and the second channel. The n in the brackets represents the fixed delay in the time domain, which can be 0, 1, 2, ..., M-1, a total of M, which is equal to the order of the filter; S 1 (m) represents the mth data of the real-time effective value sequence of the first data channel (current signal); S 2 (m+n) represents the m+nth data of the real-time effective value sequence of the voiceprint channel; Σ represents the summation symbol, whose lower limit is 0 and upper limit is NM.

[0133] The cross-correlation function sequence is obtained in this way:

[0134]

[0135] Define the cross-correlation vector:

[0136]

[0137] In the formula, the superscript T represents the transpose, that is, the cross-correlation vector is a column vector.

[0138] In the embodiment of the present application, the formula (6) defines is a real symmetric square matrix of order M, which is reversible. Find its inverse matrix and let it be Z (M) .

[0139] In the embodiment of the present application, the optimal tap weight vector is calculated:

[0140]

[0141] in, is the cross-correlation vector of the M-order first channel and the second channel defined by formula (8); Z (M)The autocorrelation matrix of the first channel and the first channel defined by formula (6) The inverse matrix is ​​an M-order square matrix; w (M) is the optimal tap weight vector of the FIR filter, which has M components.

[0142] In the embodiment of the present application, determining the first filter order in S3 includes the following steps:

[0143] By transposing the optimal tap weight vector, an estimated value of the real-time effective value of the voiceprint signal is calculated;

[0144] Calculate the mean square error based on the estimated value and the actual value;

[0145] Take the filter order M as 1, 2, 3, 4, ... respectively, and repeatedly calculate the corresponding mean square error J(M);

[0146] Select the minimum M that satisfies J(M) < 0.1% as the optimal filter order M 0 .

[0147] In the embodiment of the present application, calculating the estimated value of the real-time effective value of the voiceprint signal includes:

[0148] Assume that the optimal tap weight vector w obtained by formula (9) (M) The transpose of is:

[0149] [w 0 ,w 1 ,w 2 ,…,w M-1 ](10)

[0150] Where M is the order of the filter; w 0 Represents the 0th optimal tap weight coefficient of the filter, a real number, and so on.

[0151] In the embodiment of the present application, the estimated value of the real-time effective value of the voiceprint signal is expressed as:

[0152]

[0153] In the formula, It is expressed as the nth estimated value of the real-time effective value of the second channel under the M-order filter jump condition; w i The i-th component of the optimal tap weight vector obtained by expression (10); S 1 (ni) represents the ni-th value of the real-time effective value sequence of the first channel defined by formula (2); M is the filter order, a positive integer; Σ represents the summation, with a lower limit of 0 and an upper limit of M-1.

[0154] In the embodiment of the present application, the mean square error between the real-time effective value and the estimated value of the voiceprint signal is expressed as:

[0155]

[0156] Wherein, J(M) represents the cost function when the filter order is M, where the mean square error between the estimated value and the true value is taken; N represents the length of the current real-time effective value sequence and the voiceprint real-time effective value sequence; M represents the filter order; Calculated by formula (11), it represents the ith estimated value of the real-time effective value of the voiceprint; S 2 (i) is calculated by formula (4), which represents the i-th value of the real-time effective value of the voiceprint; Σ represents the summation symbol, whose lower limit is N-M+2 and upper limit is N. Formula (12) obtains the mean square error between the estimated value and the true value of the real-time effective value of the voiceprint signal estimated by the optimal tap weight coefficient of the filter when the filter order is M, which is used to evaluate the performance of the filter.

[0157] In the embodiment of the present application, determining the filter order M includes:

[0158] The order M of the filter is taken as 1, 2, 3, 4, ..., and the corresponding mean square error J(M) is calculated according to formulas (5), (6), (7), (8), (9), (10), (11), and (12). In principle, the larger the filter order, the smaller the mean square error J(M). However, the larger the order, the greater the computing power overhead of the server. Considering the limitations of the server computing power and the error allowed by the actual project, when J(M) < 0.1%, the order M at this time is determined to be the filter order. The process box is as follows Figure 4 As shown, the J(M) curve is Figure 5 shown.

[0159] It should be noted that the filter order that meets the error performance is tested by calculating the mean square error during modeling calculation. Once the modeling calculation is completed, the filter order is determined. In the subsequent monitoring operation process, this order can be used as the order of the real-time model, without repeated calculations, saving server computing power.

[0160] In the embodiment of the present application, S4 includes the following steps:

[0161] Assume that the order obtained according to S4 is M 0 , the optimal tap weight vector obtained according to formulas (5), (6), (7), (8), (9), and (10) is:

[0162]

[0163] This vector has M 0 components, corresponding to M 0 A point in a dimensional space is denoted by

[0164]

[0165] According to the description of data collection in S1, point O is the average model of the 72-hour continuous operation period after the equipment overhaul.

[0166] For S described by formula (2), 1 and S described by formula (4) 2 , divided by hour, 72 sets of data are obtained, and the optimal weight vectors are obtained according to (5), (6), (7), (8), (9), and (10) and mapped to M 0 space, and get the point sequence: P 0 ,P 1 ,P 2 ,…,P 71 , these 72 points correspond to the dynamic model for each hour of 72 hours of continuous operation after overhaul.

[0167] In M 0 In space, let the maximum distance from O to these 72 points be r, and make M with O as the center and r as the radius. 0 The corresponding score for the surface of the ball is 95 points, and the corresponding score for the center of the ball is 100. The specific process is as follows Figure 6 shown.

[0168] In the embodiment of the present application, for the data collected in any hour during the operation, the optimal tap weight vector is obtained according to formulas (1), (2), (3), (4), (5), (6), (7), (8), (9), and (10), and the distance d between the point corresponding to the vector in the n-dimensional space and the center of the sphere O corresponding to the model determined in 8 is calculated, and the score is calculated according to the following formula:

[0169]

[0170] In the formula, socre represents the score, d represents the distance between the point corresponding to the current optimal tap weight vector in the n-dimensional space and the sphere center O corresponding to the model determined by S5; r represents the radius of the sphere corresponding to the model determined by S5. The specific process is as follows Figure 7 As shown, the real-time score curve is as follows Figure 8 As shown, the 426th hour score dropped instantly to 68 points because other equipment on the same busbar as the equipment's power supply was faulty and grounded, causing an instantaneous imbalance in the motor power supply voltage. After the faulty equipment protection was activated and the faulty equipment was removed, the system returned to normal.

[0171] It should be noted that filters are originally used to achieve data smoothing, filtering, prediction and other functions. This method uses the mapping of filter parameters (optimal tap weight vector) in multidimensional space as a digital model of the device. Figure 2In the real scene shown, the real device inputs current and outputs the voiceprint signal at the key position. This input-output relationship is equivalent to a digital filter, and the filter parameters are the digital model of the device. Furthermore, the digital model is mapped to a multi-dimensional space, and the spatial distance is used to express the degree to which the current state of the device deviates from the original intact state, making it easy to compare and sort models, which is conducive to simple and intuitive grasp by operators.

[0172] It should be noted that the current signal is the excitation of the device, and the soundprint signal at the key position is the response of the device. From the perspective of signal observation, the role of the device is to transform the excitation into the response. When the device maintains its initial intact and healthy state, this transformation is fixed. From a mathematical point of view, the function that describes this transformation is fixed. When the health state of the device changes, the corresponding transformation changes, and the corresponding function also changes. Therefore, mathematical modeling is needed to equivalently transform the excitation-response.

[0173] In the embodiment of the present application, the scoring mechanism in S5 includes: 80 points or more indicate that the equipment is completely normal and does not require attention; 60 to 80 points indicate that the equipment can still operate, but there are potential problems and need to be paid more attention. If there is an opportunity for downtime maintenance, maintenance should be arranged; below 60 points indicate that the equipment is on the verge of failure and downtime maintenance should be arranged as soon as possible.

[0174] In the embodiments of the present application, Figure 3 As shown, the data collected by the first channel (current signal) of the synchronous acquisition card is converted into the current real-time effective value sequence S by formula (2): 1 (0),S 1 (1),

[0175] S 1 (2),… as input, the data collected by the second channel (voiceprint signal) of the synchronous acquisition card is passed through formula (4) to obtain the real-time effective value sequence S of the voiceprint 2 (0),S 2 (1),S 2 (2),…as the expected response. The precise standard for evaluating this mathematical model is the error e(n) between the model output and the expected response. e(n) itself is a random variable and is not a good performance indicator. The mean square error between the expected response and the output is used as the evaluation indicator. The purpose is to design an algorithm-implementable mathematical model that minimizes the mean square error between the expected response and the output. The parameter set w of this mathematical model is 0 ,w 1 ,w 2 ,…that is, a digital model of the device’s response to current input and voiceprint.

[0176] It should be noted that, considering Figure 3 The mathematical model shown is equivalent to an IIR (infinite impulse response) filter. The output y(n) of the filter in discrete time is the convolution of the filter input and the filter impulse response, that is:

[0177]

[0178] Where y(n) represents the discrete-time output of the IIR filter; w k represents the kth tap weight coefficient of the IIR filter; S 1 (nk) represents the nkth value of the filter input signal (real-time effective value of current); Σ represents the summation symbol, with a lower limit of 0 and an upper limit of infinity.

[0179] It should be noted that the error between the expected output and the output is:

[0180] e(n)=S 2 (n)-y(n) (15)

[0181] Where e(n) represents the error between the expected output and the filter output; S 2 (n) represents the expected output, that is, the real-time effective value of the second channel (voiceprint signal) of the acquisition card; y(n) represents the output of the filter.

[0182] It should be noted that the minimum mean square error is used to define the cost function:

[0183] J(n)=E{e 2 (n)}=E{(S 2 (n)-y(n)) 2} (16)

[0184] In the formula, J(n) represents the cost function; E{} represents the mathematical expectation; e 2 (n) represents the square of the error; S 2 (n)-y(n)) 2 Represents the square of the difference between the desired output and the filter output.

[0185] It should be noted that the gradient operator is defined as:

[0186]

[0187] In the formula, Indicates w k Find the partial derivative.

[0188] Function J(n) for weight parameter w k The partial derivative of is expressed as:

[0189]

[0190] In the formula, represents the gradient operator operation defined in formula (17) on J(n); It means to find the value of J(n) for w k The partial derivative of ; J(n) represents the cost function defined in equation (16).

[0191] From formulas (16) and (18), we can get:

[0192]

[0193] Substituting formula (14) into formula (19), we get:

[0194]

[0195] In the formula, represents the gradient operator operation defined by formula (17) on J(n); E{} represents the mathematical expectation; e(n) represents the error between the expected output and the filter output; S 1 (nk) represents the nkth value of the filter input sequence (i.e., the real-time effective value sequence of the current). Let e opt (n) represents the optimal parameter set w for selecting the filter 0 , w 1 , w 2 ,…after that, the error is obtained.

[0196] According to formula (20), it should satisfy:

[0197] E{e opt (n)S 1 (nk)}=0k=0,1,2, (21)

[0198] In the formula, E{} represents the mathematical expectation; e opt (n) represents the optimal parameter set w for selecting the filter 0 , w 1 , w 2 , ..., the error obtained; S 1 (nk) represents the nkth value of the filter input sequence (i.e., the current real-time effective value sequence). Substituting formula (15) into formula (21), we get:

[0199]

[0200] In the formula, E{} represents the mathematical expectation; S 1 (nk) represents the nkth value of the filter input sequence (i.e., the current real-time effective value sequence); S 2(n) represents the expected response of the filter, i.e., the nth value of the voiceprint effective value sequence; w opt,i represents the i-th parameter of the optimal parameter group of the filter; S 1 (ni) represents the ni-th value of the filter input sequence (i.e., the current real-time effective value sequence); ∑ represents the summation symbol, with a lower limit of 0 and an upper limit of infinity.

[0201] Expand formula (22) and rearrange it to obtain:

[0202]

[0203] In the formula, E{S 1 (nk)S 1 (ni)} is the autocorrelation function R of the filter input at lag ik 1,1 (ik), that is:

[0204] R 1,1 (ik)=E{S 1 (nk)S 1 (ni)} (24)

[0205] Among them, E{S 1 (nk)S 2 (n)} is the cross-correlation function R between the filter input and the expected response at lag -k 1,2 (-k), that is:

[0206] R 1,2 (-k) = E{S 1 (nk)S 2 (n)} (25)

[0207] Using formulas (24) and (25), formula (23) can be simplified as:

[0208]

[0209] It should be noted that formula (26) describes the conditions that the optimal filter parameter set must satisfy. The real-time effective value sequence of the current is used as the input signal, and the effective value sequence of the voiceprint is used as the expected output signal. Both are directly measured and calculated by the synchronous acquisition card. Therefore, the autocorrelation function of the input signal, i.e., R 1,1 (τ) can be estimated; the cross-correlation function between the input signal and the expected output signal is R 1,2(τ) can be estimated, so the optimal filter parameter set can be obtained by solving equation group (26). However, equation group (26) requires solving an infinite number of equations, which is mathematically impossible to handle. A realistic idea is to limit the number of equations in equation group (26), that is, to use a filter with a finite number of parameters to approximate a filter composed of infinite parameters, as long as the error is not greater than the error actually allowed in the engineering.

[0210] In the embodiment of the present application, the IIR filter is changed to a FIR (finite impulse response) filter, and the number of taps of the FIR filter is assumed to be M, that is, its tap weight coefficient group is: w 0 ,w 1 ,w 2 ,…,w M-1 , formula (26) is approximately:

[0211]

[0212] In the formula, w opt,i represents the i-th optimal tap weight coefficient of the FIR filter; R 1,1 (ik) represents the ikth value of the autocorrelation function of the real-time effective value sequence of the current; R 1,2 (-k) represents the -kth value of the cross-correlation function between the real-time effective value sequence of the current and the real-time effective value sequence of the voiceprint; Σ represents the summation, with a lower limit of 0 and an upper limit of M-1.

[0213] In the embodiment of the present application, the output of the FIR filter is:

[0214]

[0215] Where y(n) represents the discrete time domain sequence output by the FIR filter; w opt,i represents the i-th optimal tap weight coefficient of the FIR filter; S 1 (ni) represents the ni-th value of the real-time current effective value sequence; Σ represents the summation, with a lower limit of 0 and an upper limit of M-1.

[0216] In the embodiment of the present application, the autocorrelation matrix defined by formula (6) is and the cross-correlation vector defined by formula (8) The equation group (27) can be expressed in matrix form as:

[0217]

[0218] In the formula, w (M) =[w opt,0 ,w opt,1 ,w opt,2 ,…,w opt,M-1] T , that is: a column vector consisting of M optimal tap weight coefficients.

[0219] In the embodiment of the present application, multiply both sides of formula (29) by The inverse matrix of (M) , we get formula (9). Formula (9) describes the method of obtaining the optimal tap weight coefficients of the M-order FIR filter.

[0220] In the embodiment of the present application, for the determination of the order M, the mean square error between the filter output and the expected output is used as the performance evaluation, and the function J(M) of the mean square error with respect to the order M is obtained by using formula (12). The order M is calculated continuously from 1 and J(M) is calculated until the design index (not more than 0.1%) is met, and the order M is determined.

[0221] In the embodiment of the present application, the data of continuous operation for 72 hours after the equipment overhaul is used as the equipment modeling data. At this time, the equipment is considered to be in good condition. The order of the FIR filter obtained by modeling is set to M, and the optimal tap weight coefficient is: w opt,0 ,w opt,1 ,w opt,2 ,…,w opt,M-1 . This tap weight coefficient is the digital model of the equipment in good condition, corresponding to a point in the M-dimensional space, called the "good point". In the subsequent operation of the equipment, the closer the point where the real digital model of the equipment is mapped to the M-dimensional space is to the "good point", the healthier the equipment is. In order to map this distance with the percentage score that represents the health level of the equipment, and to consider the error redundancy caused by the necessary noise signal, the optimal tap weight coefficient w is obtained by modeling the equipment after 72 hours of continuous operation after overhaul. opt,0 ,w opt,1 ,w opt,2 ,…,remember w opt,M-1 The corresponding point in the M-dimensional space is O, and the value of this point is assigned to 100. The 72 hours of continuous operation data after the equipment overhaul is evenly divided into 72 segments, each segment corresponds to one hour of data, and 72 sets of optimal tap weight coefficients w are modeled separately. opt,0 ,w opt,1 ,w opt,2 ,…,w opt,M-1 , corresponding to 72 points in M-dimensional space: P 1 ,P 2 ,P 3 ,…,P 72 .make:

[0222] r=max{|OP i |,1≤i≤72,i∈N} (30)

[0223] In the formula, max{} represents the maximum value of the elements listed in the curly brackets; |OP i | represents point O to point P i distance; 1≤i≤72, i∈N represents all integers from 1 to 72.

[0224] In the embodiment of the present application, the M-dimensional sphere with O as the center and r as the radius is assigned a score of 95. The score of any point in the M-dimensional space is determined by the distance d from the center O according to formula (13). Formula (13) is essentially a linear transformation of the distance to the score. Any point in this M-dimensional space is a dynamic digital model of the device, which is obtained by modeling the data of every hour of real-time operation.

[0225] Example 3

[0226] The above is a schematic scheme of a method for early warning of induced draft fan faults in a power plant based on current-soundprint in this embodiment. It should be noted that the technical scheme of the system for early warning of induced draft fan faults in a power plant based on current-soundprint and the technical scheme of the method for early warning of induced draft fan faults in a power plant based on current-soundprint belong to the same concept. For details not described in detail in the technical scheme of the system for early warning of induced draft fan faults in a power plant based on current-soundprint in this embodiment, please refer to the description of the technical scheme of the method for early warning of induced draft fan faults in a power plant based on current-soundprint.

[0227] This embodiment also provides a power plant induced draft fan fault warning system based on current-voiceprint, including a data acquisition module, a solution module, a filter order determination module, a calculation module and a generation module;

[0228] In the embodiment of the present application, the data acquisition module includes collecting first data and second data of the induced draft fan;

[0229] In the embodiment of the present application, the data acquisition and preprocessing module includes collecting first data and second data, the first data is the current signal of the induced draft fan driving motor, and the second data is the sound print signal of the key position.

[0230] In the embodiment of the present application, the solution module includes calculating the autocorrelation matrix and the cross-correlation vector of the two sets of preprocessed data to solve the tap weight vector of the first filter;

[0231] In an embodiment of the present application, preprocessing of the collected data includes anti-aliasing filtering to prevent frequency aliasing, and calculating the effective value of each channel signal, so as to obtain a preprocessed effective value sequence of the current and voiceprint signals, providing an accurate data basis for subsequent modeling.

[0232] In the embodiment of the present application, the filter order determination module includes solving the mean square error between the real-time effective value and the estimated value of the second data based on the tap weight vector to determine the first filter order;

[0233] In the embodiment of the present application, based on the result of the mean square error, the influence of different orders M on the filter performance is analyzed, and the limitation of server computing power and the error allowed by the actual project are comprehensively considered to determine the most suitable filter order M.

[0234] In the embodiment of the present application, the calculation module includes combining the first filter order with the weight vector, calculating the deviation and converting it into a component value;

[0235] In an embodiment of the present application, the generation module includes generating a device health value based on the score and in combination with a scoring mechanism.

[0236] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for early warning of induced draft fan failure in a power plant based on current-soundprint, characterized in that: include: Collecting first data and second data of the induced draft fan; The autocorrelation matrix and the cross-correlation vector are calculated for the two sets of preprocessed data to solve the tap weight vector of the first filter; Based on the tap weight vector, solving the mean square error between the real-time effective value of the second data and the estimated value, and determining the order of the first filter; The first filter order is combined with the weight vector, the deviation is calculated and converted into a component value; Based on the score and combined with the scoring mechanism, a device health value is generated.

2. The method for early warning of induced draft fan failure in a power plant based on current-soundprint according to claim 1, characterized in that: The preprocessing comprises: Perform anti-aliasing filtering on the data of all channels; Assume that the first channel collects the first data, and the second channel collects the second data; The first data is a current signal of the induced draft fan driving motor, and the second data is a sound print signal of a key position; Based on the current signal and the voiceprint signal after anti-aliasing filtering, the effective value of each channel signal is calculated to obtain a first data effective value sequence and a second data effective value sequence.

3. The power plant induced draft fan fault early warning method based on current-soundprint according to claim 1 or 2, characterized in that: The second-order statistics estimation comprises: Calculate the M-order autocorrelation matrix of the first data effective value sequence; An M-th order cross-correlation vector between the first data and the second data is calculated.

4. The method for early warning of induced draft fan failure in a power plant based on current-soundprint according to claim 3, characterized in that: The step of calculating the M-order autocorrelation matrix of the first data effective value sequence includes: For a given positive integer order M, calculate the autocorrelation function; Based on the results of the autocorrelation function, an M-order autocorrelation matrix is ​​constructed; The calculating the M-order cross-correlation vector between the first data and the second data comprises: For a given positive integer order M, calculate the cross-correlation function; According to the result of the cross-correlation function, an M-order cross-correlation function sequence is obtained; The cross-correlation function sequence is combined into a column vector, namely, the M-order cross-correlation vector.

5. The method for early warning of induced draft fan failure in a power plant based on current-soundprint according to claim 4, characterized in that: The step of solving the tap weight vector of the first filter comprises: Based on the M-order autocorrelation matrix of the first data effective value sequence, calculating the inverse matrix of the autocorrelation matrix; The tap weight vector is calculated by applying the inverse matrix of the autocorrelation matrix to the M-order cross-correlation vector; Based on the tap weight vector, the tap weight vector that minimizes the mean square error is selected as the optimal tap weight vector.

6. The method for early warning of induced draft fan failure in a power plant based on current-soundprint according to claim 5, characterized in that: The step of determining the first filter order by solving the mean square error between the real-time effective value of the second data and the estimated value comprises: Calculate an estimated value of the real-time effective value of the second data by transposing the optimal tap weight vector; Calculate the mean square error between the real-time effective value of the second data and the estimated value; It is determined whether the mean square error is less than a first threshold value. If so, the mean square error is the optimal order of the first filter.

7. The method for early warning of induced draft fan failure in a power plant based on current-soundprint according to claim 6, characterized in that: The step of combining the first filter order with the weight vector, calculating the deviation and converting it into a component value comprises the following steps: Based on the optimal order of the first filter and in combination with the optimal tap weight vector, mapping is performed to obtain the average model point within a preset time period; Based on the optimal order of the first filter and in combination with the optimal weight vector, the dynamic model sequence points within the preset time step are mapped; For each dynamic model point in the dynamic model sequence points, the Euclidean distance between each dynamic model point and the average model point is calculated, and the maximum Euclidean distance from all dynamic model points to the average model point is determined; For the new real-time monitoring data collected within the preset time step, the corresponding optimal tap weight vector is calculated and mapped to obtain the real-time monitoring data mapping points; Compare the real-time monitoring data mapping point with the average model point to obtain the Euclidean distance between the real-time monitoring data mapping point and the average model point; Based on the maximum Euclidean distance from all dynamic model points to the average model point and the Euclidean distance between the real-time monitoring data mapping point and the average model point, it is converted into a score according to the scoring formula.

8. A system based on the current-soundprint based power plant induced draft fan fault early warning method according to claim 1, characterized in that: A data acquisition module, used for acquiring first data and second data of the induced draft fan; A solution module, used for calculating the autocorrelation matrix and the cross-correlation vector of the two sets of preprocessed data, and solving the tap weight vector of the first filter; A filter order determination module, used to solve the mean square error between the real-time effective value and the estimated value of the second data based on the tap weight vector, and determine the first filter order; A calculation module, used for combining the first filter order with the weight vector, calculating the deviation and converting it into a component value; The generation module is used to generate a device health value based on the score and in combination with the scoring mechanism.

9. A computing device, characterized in that include: Memory, used to store programs; A processor is used to load the program to execute the steps of the power plant induced draft fan fault early warning method based on current-voiceprint as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a program, characterized in that: When the program is executed by the processor, the steps of the power plant induced draft fan fault early warning method based on current-voiceprint as described in any one of claims 1 to 7 are implemented.

Images