Fault diagnosis method and device for analog circuit of hemodialysis equipment based on BP neural network

Abstract

The present application relates to a method and device for diagnosing faults in analog circuits of hemodialysis equipment based on a BP neural network. The method comprises: obtaining an input signal and processing it to obtain a spectrum analysis; calculating the power spectrum density of the signal, analyzing the energy distribution of the signal at different frequencies to determine whether the faulty circuit is a high-frequency or low-frequency signal output circuit; if it is the former, analyzing the frequency magnitude of its energy concentration area to determine the corresponding circuit and obtain the fault characteristics; if it is the latter, setting a frequency threshold to distinguish between high-frequency and low-frequency areas, if the energy in the low-frequency area is higher than that in the high-frequency area, performing multi-resolution analysis on the input signal to obtain the fault characteristics; if the energy in the low-frequency area is not higher than that in the high-frequency area, performing wavelet packet analysis to obtain the fault characteristics; analyzing the fault characteristics to obtain the fault analysis results. The present application has the advantages of combining the advantages of wavelet transform and artificial neural network, and optimizing the extraction and identification of circuit fault characteristics.

Description

Technical Field

[0001] The present application relates to the field of hemodialysis equipment, and in particular to a method and device for diagnosing faults in analog circuits of hemodialysis equipment based on a BP neural network. Background Art

[0002] With the continuous progress of modern science and technology, electronic equipment has been widely used in many fields such as communication, household, automatic control, measurement, scientific research, transportation, etc. At the same time, with the update and development of semiconductor and integrated circuit technology, the structural scale of circuit systems is also expanding rapidly. However, this complex circuit structure increases the probability and complexity of failure. Literature research shows that in many modern equipment, especially modern weapons and equipment, the maintenance cost may account for more than 50% of the overall cost, and the maintenance cost of airborne electronic systems is as high as 67%. Although the area occupied by analog circuits in integrated circuits is very small, not exceeding 5%, its related costs can account for more than 95% of the entire integrated circuit cost. Therefore, if the integrated circuit industry wants to achieve a technological breakthrough, it must give priority to solving the key technical bottleneck of analog circuit detection.

[0003] Artificial Neural Network (ANN), referred to as neural network, is an abstract mathematical model based on modern neuroscience, which aims to simulate the functions of the human brain. After years of research and development, artificial neural networks have formed a variety of network models, such as multilayer perceptron, Hopfield network, ART network, adaptive resonance theory, probabilistic neural network and RBF network. These networks are suitable for different application fields due to their structural differences. Neural networks have been widely used in analog circuit fault diagnosis in the late 1980s. With the development of theoretical technology and the verification of a large number of practical applications, neural networks have gradually become an important development trend. Among the many neural network methods, BP neural network is particularly suitable for the field of circuit fault diagnosis due to its excellent pattern classification ability.

[0004] In the process of circuit fault diagnosis, in order to achieve effective fault identification and classification, the neural network model needs to accurately input the feature vector that can characterize the characteristic information of the analog circuit. As a time-frequency analysis method, wavelet transform can extract the information that characterizes the characteristics of the analog circuit by decomposing the signal to form a feature vector, thereby supporting the fault diagnosis and classification of the circuit. Wavelet transform has good sensitivity, less computing requirements and strong noise suppression ability, and does not require a specific mathematical model. However, due to the large time domain width of the circuit, wavelet transform may cause time delay problems in some cases, affecting the accuracy of the analysis. In addition, the choice of wavelet function will also affect the detection results. Summary of the invention

[0005] In order to combine the advantages of wavelet transform and artificial neural network and optimize the extraction and identification of circuit fault features, the present application provides a method and device for diagnosing faults in analog circuits of hemodialysis equipment based on BP neural network.

[0006] In the first aspect, the present application provides a method for diagnosing faults in analog circuits of hemodialysis equipment based on a BP neural network, which adopts the following technical solution:

[0007] A method for diagnosing a fault in a hemodialysis device analog circuit based on a BP neural network comprises the following steps:

[0008] S1. Get the input signal and process it to get spectrum analysis;

[0009] S2. Calculate the power spectrum density of the signal and analyze the energy distribution of the signal at different frequencies to determine whether the faulty circuit is a high-frequency signal output circuit or a low-frequency signal output circuit;

[0010] S3. If it is a high-frequency signal output circuit, the frequency magnitude of its energy concentration area is analyzed to determine the corresponding circuit and obtain the fault characteristics; wherein the corresponding circuit includes a bubble detection circuit and a heart rate monitoring circuit, and the energy concentration area is a frequency area corresponding to the energy portion exceeding a preset ratio;

[0011] If it is a low-frequency signal output circuit, a frequency threshold is set to distinguish between the high-frequency area and the low-frequency area. If the energy of the low-frequency area is higher than that of the high-frequency area, a multi-resolution analysis is performed on the input signal to obtain the fault characteristics; if the energy of the low-frequency area is not higher than that of the high-frequency area, a wavelet packet analysis is performed to obtain the fault characteristics;

[0012] S4. Analyze the fault characteristics and obtain the fault analysis results.

[0013] Optionally, the S1 includes:

[0014] S11. Select signal collection points and connect;

[0015] S12. Perform signal acquisition;

[0016] S13. Performing a fast Fourier transform on the collected signal to obtain a spectrum distribution;

[0017] S14. Extract frequency components and corresponding amplitude information.

[0018] Optionally, S2 includes:

[0019] S21. Calculate the power spectral density using the Welch method;

[0020] S22. Calculate the total energy of the high frequency region and the low frequency region respectively according to the calculated power spectrum density curve, wherein the frequency range of the high frequency region and the frequency range of the low frequency region are preset;

[0021] S23. Calculate the energy ratio of the high-frequency area and the low-frequency area. If the energy ratio is greater than the preset threshold, it is determined that the signal is mainly composed of high-frequency energy and the fault circuit is a high-frequency signal output circuit; if the energy ratio is less than the preset threshold, it is determined that the energy is mainly composed of low-frequency energy and the fault circuit is a low-frequency signal output circuit.

[0022] Optionally, the S4 includes:

[0023] The fault characteristics of the high-frequency signal output circuit are input into the pre-trained BP neural network of the corresponding specific circuit to obtain a fault analysis result, wherein the pre-trained BP neural network of the specific circuit is trained based on a sample set of fault characteristic vectors of the high-frequency signal output circuit with calibrated fault types to output a fault mode.

[0024] Optionally, the S4 includes:

[0025] The fault characteristics of the low-frequency signal output circuit are input into the pre-trained BP neural network of the corresponding general-purpose circuit to obtain the fault analysis result, wherein the pre-trained BP neural network is trained according to a sample set of fault characteristic vectors of the low-frequency signal output circuit with pre-calibrated fault types.

[0026] Optionally, the training steps of the pre-trained BP neural network of the specific circuit include:

[0027] S41. Perform analog circuit simulation to determine the components that may affect the circuit, and then determine the type of fault that may occur;

[0028] S42. Extracting a fault feature vector, wherein the fault mode and the fault feature vector have a clear correspondence with each other;

[0029] S43. Processing the fault feature vector based on the normalization method to form the input and output sample sets required for neural network training;

[0030] S44. Build and train neural networks.

[0031] Optionally, S44 includes:

[0032] S441. Setting the basic structure of the BP neural network, the basic structure includes the number of nodes in the input layer, the hidden layer and the output layer;

[0033] S442. Use the error back propagation algorithm for training to minimize the prediction error by continuously adjusting the network weights;

[0034] S443. Input the training samples into the neural network, and continuously adjust the weights through the back propagation algorithm to gradually reduce the output error until the set error target is reached;

[0035] S444. After the training is completed, the test samples are input into the neural network for performance testing to verify the accuracy of the network's fault classification of unseen data.

[0036] In the second aspect, the present application provides a hemodialysis equipment analog circuit fault diagnosis device based on BP neural network, which adopts the following technical solution:

[0037] A BP neural network-based hemodialysis equipment analog circuit fault diagnosis device, comprising:

[0038] A spectrum analysis module, used to obtain input signals and process them to obtain spectrum analysis;

[0039] A fault circuit classification module is used to calculate the power spectrum density of the signal and analyze the energy distribution of the signal at different frequencies to determine whether the fault circuit is a high-frequency signal output circuit or a low-frequency signal output circuit;

[0040] A fault feature extraction module is used to analyze the frequency magnitude of the energy concentration area of ​​the high-frequency signal output circuit to determine the corresponding circuit and obtain the fault feature; wherein the corresponding circuit includes a bubble detection circuit and a heart rate monitoring circuit, and the energy concentration area is a frequency area corresponding to the energy portion exceeding a preset ratio;

[0041] If it is a low-frequency signal output circuit, a frequency threshold is set to distinguish between the high-frequency area and the low-frequency area. If the energy of the low-frequency area is higher than that of the high-frequency area, a multi-resolution analysis is performed on the input signal to obtain the fault characteristics; if the energy of the low-frequency area is not higher than that of the high-frequency area, a wavelet packet analysis is performed to obtain the fault characteristics;

[0042] The analysis module is used to analyze the fault characteristics and obtain the fault analysis results.

[0043] In a third aspect, the present application provides a computer device, which adopts the following technical solution:

[0044] A computer device comprising:

[0045] one or more processors;

[0046] Memory;

[0047] One or more applications, wherein the one or more applications are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs are configured to: execute the above-mentioned BP neural network-based hemodialysis equipment simulation circuit fault diagnosis method.

[0048] In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution:

[0049] A computer-readable storage medium stores a computer program that can be loaded by a processor and execute the above method.

[0050] The storage medium stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement: the above-mentioned BP neural network-based hemodialysis equipment simulation circuit fault diagnosis method.

[0051] In summary, the present application includes at least one of the following beneficial technical effects:

[0052] 1. By decomposing and detecting the signal step by step, we can obtain prior conditions to predict the type of faulty circuit, so as to select a BP neural network model trained on a specific circuit with better effect, or use multi-resolution analysis or wavelet packet analysis to decompose the signal in subsequent steps to obtain more and more accurate fault features, and then input them into the BP neural network model trained on a general circuit to obtain the fault type more accurately.

[0053] 2. Through the characteristics of wavelet packet transform, preprocessing the signal with wavelet packet can avoid the large scale and slow convergence speed problems caused by the application of BP network for fault diagnosis. In addition, wavelet packet transform can also effectively eliminate and delete redundant information in the signal, complete the task of reducing the neural network structure and reducing the training time.

[0054] 3. Both BP neural network and wavelet neural network can effectively perform analog circuit fault diagnosis, but in terms of convergence rate and learning ability, the wavelet neural network fault diagnosis method is significantly better than the single BP neural network method. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A flowchart of a method for diagnosing a fault in a hemodialysis device analog circuit based on a BP neural network in one embodiment of the present invention is shown.

[0056] Figure 2 A flowchart of sub-step S1 in one embodiment of the present invention is shown.

[0057] Figure 3A flowchart of sub-step S2 in one embodiment of the present invention is shown.

[0058] Figure 4 A four-layer multi-resolution decomposition tree structure diagram used for example in one embodiment of the present invention is shown.

[0059] Figure 5 FIG. 4 is a diagram showing a wavelet packet decomposition tree used for an example in an embodiment of the present invention.

[0060] Figure 6 A schematic diagram of a neural network method according to an embodiment of the present invention is shown.

[0061] Figure 7 A flowchart of sub-step S4 in one embodiment of the present invention is shown.

[0062] Figure 8 A flowchart of sub-step S44 in one embodiment of the present invention is shown.

[0063] Fig. 9 A schematic diagram of a computer device according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0064] The present application is further described in detail below in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0065] In the following description, for the purpose of explanation, many specific details are set forth in order to provide a thorough understanding of the inventive concept. As part of this specification, some of the drawings in the drawings of the present disclosure represent structures and devices in the form of block diagrams to avoid making the disclosed principles complicated and difficult to understand. For the sake of clarity, not all features of the actual implementation are necessary to be described. In addition, the language used in this disclosure has been selected primarily for readability and instructional purposes, and may not be selected to delineate or limit the subject matter of the present invention, thereby resorting to the necessary claims to determine such inventive subject matter. References to "a specific implementation" or "specific implementations" in this disclosure mean that the specific features, structures or characteristics described in conjunction with the specific implementation are included in at least one specific implementation, and multiple references to "a specific implementation" or "specific implementations" should not be understood as necessarily referring to the same specific implementation.

[0066] Unless expressly limited, the terms "a", "an", and "the" are not intended to refer to a singular entity, but rather to include general categories of which specific examples may be used for illustration. Thus, the use of the terms "a" or "an" may mean any number of at least one, including "one", "one or more", "at least one", and "one or more than one". The term "or" means any of the alternatives and any combination of the alternatives, including all alternatives, unless the alternatives are expressly indicated to be mutually exclusive. The phrase "at least one of" when combined with a list of items refers to a single item in the list or any combination of the items in the list. The phrase does not require all of the listed items unless expressly limited to that.

[0067] When an analog circuit fails to achieve the functions and goals originally envisioned during the design process, it means that a failure has occurred. The cause of the failure is relatively simple. From the entire process from production to use, it mainly comes from three aspects: circuit design, factory manufacturing, and final use. Early initial failures are usually caused by factors such as inherent defects in production, while wear failures are caused by factors such as the inevitable wear of components during the continuous use of analog circuits.

[0068] When an analog circuit fails, it often causes certain changes compared to the normal state. In order to be able to significantly express this change, people have proposed the concept of fault characteristics. The manifestation of analog circuit fault characteristics is diverse, which can be expressed as the input or output signal of the analog circuit under different conditions, or as the AC or DC voltage at a specific node of the entire circuit system.

[0069] By simulating the number of faults that may exist in the circuit at the same time, circuit faults can be divided into single faults and multiple faults.

[0070] Single fault: refers to the situation where only a single electronic component in the entire circuit fails.

[0071] Multiple faults: refers to the situation where two or more electronic components in the entire circuit fail at the same time.

[0072] Therefore, all possible situations of the circuit under test can be formed into a state space, and then the fault characteristics can be extracted through certain methods to form a feature vector space. , it must also have a given characteristic y, and there must be a mapping On the contrary, every specific system feature must have a specific state corresponding to it, that is, there is also a mapping If f and g are bijective functions, then the feature space and state space are full mappings with a one-to-one correspondence, that is, the state of the circuit system can be determined through the feature vector. Therefore, the mapping existing in the circuit fault is an important theoretical basis for judging what kind of fault the circuit system has.

[0073] Reference Figure 1 The embodiment of the present application discloses a method for diagnosing faults in analog circuits of hemodialysis equipment based on a BP neural network. The method includes the following steps S1-S4.

[0074] S1. Get the input signal and process it to get the spectrum analysis.

[0075] Specifically, refer to Figure 2 In one embodiment disclosed in the present application, S1 includes S11-S15.

[0076] S11. Select the signal collection point and connect it.

[0077] Typically, the output of an analog circuit is the only available signal acquisition point.

[0078] S12. Perform signal acquisition.

[0079] In the process of fault diagnosis of analog circuits, applying appropriate excitation signals can significantly improve the accuracy of fault feature extraction. Usually, the circuit's working signal or other input signal is selected as the excitation signal, such as pulse signals, sinusoidal signals of multiple frequencies, piecewise linear function signals, and square wave signals. The main reason why these signal types are selected is that they can trigger the typical response characteristics of the circuit, thereby showing more significant amplitude-frequency response characteristics at the output end.

[0080] Pulse signal: As a transient stimulus, pulse signal can stimulate the full-band response of the circuit. Its spectrum is relatively wide, which helps to reveal the response characteristics of the circuit at different frequencies, thereby exposing possible fault frequency bands.

[0081] Multi-frequency sine signal: Applying a multi-frequency sine signal can make the circuit respond at multiple frequencies, which is very effective for identifying abnormal characteristics within a specific frequency range. Especially in the detection of high-frequency faults, multi-frequency sine signals can provide more accurate amplitude-frequency characteristics.

[0082] Piecewise linear function signal: This type of signal has different linear characteristics in different time periods. By applying linear excitations with different characteristics in segments, the frequency response of the circuit can be further refined. This excitation method is suitable for detecting the dynamic characteristics of the circuit and its fault manifestations under different working conditions.

[0083] Square wave signal: Square wave signal contains rich harmonic components in the frequency domain, which can trigger the nonlinear characteristics in the circuit, making it easier to expose nonlinear faults that may exist in the circuit. Therefore, square wave signals have unique advantages in detecting some specific types of faults.

[0084] In order to obtain complete spectrum information, the sampling frequency setting must satisfy the Nyquist sampling theorem, that is, the sampling frequency should be at least twice the highest frequency of the signal. Too low a sampling frequency will lead to aliasing, causing high-frequency components to mix into the low-frequency area, affecting the extraction of fault features; while too high a sampling frequency will increase the amount of data processing calculations.

[0085] S13. Perform fast Fourier transform on the collected signal to obtain spectrum distribution.

[0086] S14. Extract frequency components and corresponding amplitude information.

[0087] S2. Calculate the power spectrum density of the signal and analyze the energy distribution of the signal at different frequencies to determine whether the faulty circuit is a high-frequency signal output circuit or a low-frequency signal output circuit.

[0088] Power Spectral Density (PSD) is a statistic used to describe the energy distribution of a signal in the frequency domain. It represents the distribution of the signal's power at different frequencies and is usually used to analyze the frequency characteristics of random and complex signals.

[0089] The spectrum is a representation of a signal in the frequency domain, usually obtained by Fourier transform. It shows the distribution of the amplitude or magnitude of the signal at each frequency. The power spectral density is a squared normalized representation of the spectrum, reflecting the power distribution of the signal at different frequencies. It not only tells us at which frequencies the signal has energy, but also quantifies the power intensity at these frequencies.

[0090] Specifically, in one embodiment disclosed in the present application, refer to Figure 3 , S2 includes S21-S23.

[0091] S21. Calculate the power spectral density using the Welch method.

[0092] The power spectral density is calculated using the Welch method to reduce the variance of the spectrum estimate. The Hanning window is selected with a window length of 256 points and a 50% overlap rate to smooth the power spectral density curve and reduce spectrum leakage.

[0093] S22. According to the calculated power spectrum density curve, the total energy of the high frequency region and the low frequency region are calculated respectively, wherein the frequency range of the high frequency region and the frequency range of the low frequency region are preset.

[0094] S23. Calculate the energy ratio of the high-frequency area and the low-frequency area. If the energy ratio is greater than the preset threshold, it is determined that the signal is mainly composed of high-frequency energy and the fault circuit is a high-frequency signal output circuit; if the energy ratio is less than the preset threshold, it is determined that the energy is mainly composed of low-frequency energy and the fault circuit is a low-frequency signal output circuit.

[0095] In the analog circuit part of the dialysis equipment, different circuits process different signal frequency ranges. Ultrasonic sensor interface circuits are used to detect bubbles in blood or dialysate, usually using ultrasonic technology, and the ultrasonic frequency is generally between hundreds of kilohertz (kHz) and several megahertz (MHz). In the process of signal acquisition and processing, ECG / heart rate monitoring circuits need to process higher frequency signal components such as filtering noise, detecting spikes and heart rate variability, which involves high-frequency signal processing, especially in the filtering and signal conditioning stages. Circuits such as blood flow monitoring circuits, temperature control circuits, pressure monitoring and control circuits, conductivity monitoring circuits, and physiological signal monitoring circuits, these changing signals are generally in the low frequency range, usually not more than a few hertz.

[0096] Therefore, in the embodiment of the present application, the frequency range is divided into a low frequency area and a high frequency area. For example, the frequency threshold is set according to the application scenario. (such as 100 Hz) to distinguish between the two.

[0097] Calculate the signal in the low frequency region (such as 0 to ) and high frequency regions (such as The total energy of the above:

[0098]

[0099] As an example, the preset threshold in this step can be set between 0.4 and 0.8, which can better balance the fault discrimination of the high and low frequency signal output circuits in practical applications. Of course, in practical scenarios, this value needs to be adjusted according to the specific application scenario and circuit frequency characteristics.

[0100] S3. If it is a high-frequency signal output circuit, analyze the frequency magnitude of its energy concentration area to determine the corresponding circuit and obtain the fault characteristics; wherein the corresponding circuit includes a bubble detection circuit and a heart rate monitoring circuit, and the energy concentration area is the frequency area corresponding to the energy part exceeding the preset proportion.

[0101] If it is a low-frequency signal output circuit, a frequency threshold is set to distinguish between the high-frequency area and the low-frequency area. If the energy in the low-frequency area is higher than that in the high-frequency area, multi-resolution analysis is performed on the input signal to obtain the fault characteristics; if the energy in the low-frequency area is not higher than that in the high-frequency area, wavelet packet analysis is performed to obtain the fault characteristics.

[0102] Since the frequency magnitude of the output signal of the high-frequency signal output circuit is obviously different, this can be used as a priori condition to identify the circuit. For high-frequency signals, since the signal frequency ranges processed by the analog circuit part of the bubble detection circuit and the heart rate monitoring circuit are different, there is a difference in magnitude, so the faulty circuit can be distinguished by magnitude to determine the corresponding circuit.

[0103] For low-frequency signals, the signal may be evenly distributed or unevenly distributed at different frequencies. For example, the energy of the low-frequency part of the low-frequency signal accounts for a higher proportion, and the energy of the high-frequency part accounts for a lower proportion, or vice versa. Therefore, here we compare the high and low frequency energies of the low-frequency signal twice. When the energy occupied by the low-frequency part is higher, multi-resolution analysis is used. If the frequency distribution is uniform, wavelet packet analysis is used.

[0104] Reference Figure 4 , multi-resolution analysis is to continuously decompose the signal S into two parts: approximate (low frequency) and detail (high frequency), and then further decompose the approximate part to achieve the multi-layer decomposition structure of the signal S. And its decomposition has the relationship: S=A4+D4+D3+D2+D1, where D1 is the high frequency part of S, A1 is the low frequency part of S; D2 is the high frequency part of A1, A2 is the low frequency part of A1; D3 is the high frequency part of A2, A3 is the low frequency part of A2; D4 is the high frequency part of A3, A4 is the low frequency part of A3; if it continues to decompose, it can still decompose the approximate low frequency part A4 into the approximate low frequency part A5 and the detail high frequency part D5, and the further decomposition is the same.

[0105] In addition, the preset ratio described in this step can be set according to actual conditions, such as 80%.

[0106] Compared with multi-resolution analysis, Figure 5 , wavelet packet analysis can provide a more sophisticated method for signal analysis. It can fully decompose the original signal into high-frequency details and low-frequency overviews in its subspace, and then further decompose its high-frequency details and low-frequency overviews. However, multi-resolution analysis cannot complete such decomposition. Figure 5 In the figure, D1 is the high frequency part of S, and A1 is the low frequency part of S; DA2 is the high frequency part of A1, and AA2 is the low frequency part of A1; DAA3 is the high frequency part of AA2, and AAA3 is the low frequency part of AA2; DDA3 is the high frequency part of DA2, and ADA3 is the low frequency part of DA2; DD2 is the high frequency part of D1, and AD2 is the low frequency part of D1; DAD3 is the high frequency part of AD2, and AAD3 is the low frequency part of AD2; DDD3 is the high frequency part of DD2, and ADD3 is the low frequency part of DD2.

[0107] Taking energy as an element can simulate various characteristic vectors under various fault forms. The specific steps are as follows:

[0108] 1) Perform j-layer wavelet packet decomposition on a certain fault sampling signal f(t), and respectively decompose the common The correlation coefficients of the frequencies are extracted to form a sequence, that is .

[0109] 2) Order express The reconstructed signal, express The reconstructed signal is: .

[0110] 3) Set The corresponding energy is , then: ,in Represented as the reconstructed signal The specific amplitude at each discrete point.

[0111] 4) Energy as the element Dimensional vector It contains all kinds of characteristic information of this fault mode. Since the energy is often very large, P needs to be normalized and processed before it can be used as the characteristic vector of this analog circuit fault mode.

[0112] In the actual fault feature vector extraction, the wavelet packet analysis method is used, and the number of wavelet packet decomposition layers needs to be appropriately selected. If the number of decomposition layers selected is too small, it is impossible to fully extract useful fault feature vectors. If the number of decomposition layers selected is too large, the dimension of the feature vector will become larger, which will affect the training of the neural network and reduce its diagnostic ability.

[0113] S4. Analyze the fault characteristics and obtain the fault analysis results.

[0114] Specifically, according to the judgment result in S3, refer to Figure 6 In one embodiment of the present application, S4 performs the following corresponding steps:

[0115] The fault characteristics of the high-frequency signal output circuit are input into the pre-trained BP neural network of the corresponding specific circuit to obtain a fault analysis result, wherein the pre-trained BP neural network of the specific circuit is trained based on a sample set of fault characteristic vectors of the high-frequency signal output circuit with calibrated fault types to output a fault mode.

[0116] The fault characteristics of the low-frequency signal output circuit are input into the pre-trained BP neural network of the corresponding general-purpose circuit to obtain the fault analysis result, wherein the pre-trained BP neural network is trained according to a sample set of fault characteristic vectors of the low-frequency signal output circuit with pre-calibrated fault types.

[0117] In S4, the fault features of high-frequency and low-frequency circuits are input into the specific or general pre-trained BP neural network, and these features can be used as the input layer data of the neural network. Different BP neural network structures are used in the diagnosis of high-frequency signal output circuits and low-frequency signal output circuits: the BP neural network of the specific circuit is specially trained for the high-frequency features of the specific circuit, and can better capture the fault mode of the specific circuit; the BP network of the general circuit is generalized for a wider range of low-frequency circuits and has a wider applicability.

[0118] Specifically, the generalized model needs to adapt to the characteristics of multiple fault types, and its ability to distinguish the characteristics of each fault is not as high as that of the specific model. In particular, when the fault characteristics vary greatly, the generalized model may not be able to fully learn the specific patterns of each type. The generalized model is prone to overlooking some subtle but important fault characteristics in complex scenarios, especially for high-frequency and low-frequency faults with different characteristics, and the sensitivity may be insufficient.

[0119] In addition, the generalized model contains more neurons and layers to capture multiple fault characteristics in the same model. This leads to an increase in the number of parameters, a relatively large model size, and relatively high computational requirements when performing diagnosis.

[0120] As an example, refer to Figure 6 and Figure 7 In one embodiment, the training steps of the pre-trained BP neural network of a specific circuit include S41-S44.

[0121] S41. Perform analog circuit simulation to determine the components that may affect the circuit, and further determine the type of fault that may occur.

[0122] During the simulation process, we will focus on which components can cause significant changes in circuit characteristics when a fault occurs. For example, in a high-frequency circuit, the failure of components such as capacitors and inductors may have a significant impact on the frequency or phase of the output signal. These components are considered key components because their status is directly related to the high-frequency characteristics and stability of the circuit. After determining these key components, we can focus more on the working status of these components when extracting fault characteristics.

[0123] Based on the simulation results, the possible fault types can be further clarified. Different component failures will present different changes in circuit characteristics, such as increased resistance leading to signal attenuation and capacitor failure causing frequency drift. By analyzing the simulation results, the specific fault types of specific components and their impact patterns on the circuit can be summarized.

[0124] S42. Extracting a fault feature vector, wherein the fault mode and the fault feature vector have a clear corresponding relationship with each other.

[0125] This step extracts fault feature vectors from the simulation results. These feature vectors can reflect the association between different fault modes of the circuit and signal characteristics, so that various fault types can be identified during the BP neural network training process.

[0126] By analyzing the amplitude-frequency characteristic curve of the circuit, the amplitude characteristics of the circuit at different frequency points can be obtained. These frequency response characteristics are important information for fault diagnosis, especially in high-frequency and low-frequency fault modes, where the amplitude differences at different frequency points will reflect the health of the circuit.

[0127] Through the amplitude-frequency characteristic curve of the circuit, the effective sampling point method is used to extract the signal, and the characteristic vector space is formed by preprocessing the extracted signal. This method is suitable for the situation where there are few measurable points in the circuit. As long as the sampling frequency points can be guaranteed to be sufficient, the circuit state can be correctly reflected. For example, the amplitudes at several different frequencies are selected to form the characteristic vector. The frequencies selected in practical applications are 10KHz, 15KHZ, 20KHz, 25KHz, 30KHZ, 35KHz, 40KHz, and 45KHz. AC simulation is performed on the circuits of nine fault modes respectively, and Monte-Carlo analysis is performed 7 times respectively. The amplitudes of the output points are extracted, and 63 sets of characteristic vector spaces can be obtained for training and testing of BP neural networks.

[0128] Different fault modes show different characteristics in amplitude-frequency response, so in the feature vector space, each set of vectors corresponds to a specific fault mode.

[0129] S43. Process the fault feature vector based on the normalization method to form the input and output sample sets required for neural network training.

[0130] After successfully extracting the feature vector, the extracted data vector is processed to form the input and output sample sets required for neural network training. When constructing the training neural network sample set, in order to achieve better network learning and classification effects, the feature vector is generally preprocessed by normalization method.

[0131] S44. Build and train neural networks.

[0132] As a multi-layer feedforward network, BP network is mainly composed of three main parts: input layer, output layer and hidden layer. There are many nodes representing neurons on each layer of its structure. At the same time, the propagation of data information in the network is unidirectional from the input layer to each layer, and passes through each hidden layer node in the neural network in turn, and then reaches each node of the network output layer. Assume that the BP network has n input data, which are represented by vector X respectively: .

[0133] The network will then generate m output data, which will be represented by vector Y: .

[0134] Then the corresponding neural network will have n input and m output nodes. Therefore, from another perspective, the BP neural network can be regarded as a determined nonlinear mapping relationship from an n-dimensional input space to another m-dimensional output space.

[0135] The excitation function selected by BP neural network is usually a monotonically continuously differentiable increasing function. The transfer excitation function of the hidden layer usually uses the sigmoid function or purelin pure linear function. The excitation function used here is the sigmoid function: .

[0136] The algorithm used by BP neural network for learning and training is completely different from other networks, namely the error back propagation algorithm (called BP algorithm). Its basic idea is: for q input learning samples: , it is known that the corresponding output sample is: The learning method is mainly through the actual output of the neural network And the expected output The error is used to modify the network weights so that the final actual output generated by it can be as close to the target output as possible, that is, to minimize the sum of squared errors generated by the neural network on the output layer. Moreover, in the neural network, each time the weights and deviations are changed in value, they are often proportional to the network errors, and are transmitted to each layer of the neural network through the back propagation method.

[0137] The approximate steepest descent method can be used in BP neural networks to adjust and set the weights and bias values ​​in the network:

[0138]

[0139] Since the BP neural network is a multi-layer network, each value here uses the superscript m to indicate the mth layer. It should be pointed out that: It represents the sensitivity of the network mean square error F to the change of the i-th element of the input in the m-th layer. And the sensitivity of each layer is generated by taking the partial derivative of the network mean square error F relative to the network input of that layer. It can be expressed as:

[0140]

[0141] In this formula represents the input vector of the mth layer, and express Each component of The calculation of requires the use of the chain rule because it can successfully determine the recursive relationship describing the sensitivity between the mth layer and the m+1th layer.

[0142]

[0143] From this we can see that the internal propagation of sensitivity in the neural network is through the last layer using back propagation to the first layer:

[0144]

[0145] For the recursive relation Starting point in Finally it can be expressed as:

[0146]

[0147] In summary, the BP algorithm can be summarized into three steps: the first step is to use the neural network to propagate the input data forward; the second step is to calculate the sensitivity of the last layer in the neural network. , and back-propagate the sensitivity back to the first layer; the third step is to continuously adjust the weights and bias values ​​in the network through the method of approximate steepest descent.

[0148] Specifically, refer to Figure 8 , S44 includes the following steps S441-S444.

[0149] S441. Set the basic structure of the BP neural network, which includes the number of nodes in the input layer, hidden layer and output layer.

[0150] Number of nodes in the input layer: The number of inputs to the neural network must correspond to the dimension of the fault feature vector extracted previously. For example, if the amplitude features of 8 frequency points are extracted, the number of nodes in the input layer is set to 8.

[0151] Number of nodes in the output layer: The number of nodes is often determined by the user's needs. If there are 4 failure modes, the output layer is set to 4 nodes.

[0152] Number of network layers: The number of hidden layers of the BP neural network can vary, but in most cases a 3-layer model can be used to solve the problem.

[0153] Hidden layer: The number of nodes in the hidden layer is often selected by an empirical formula, such as the empirical formula M = √(number of input layer nodes × number of output layer nodes) + α, where α is a constant. 19 nodes can be selected, and the commonly used activation function is the sigmoid function.

[0154] S442. Use the error back propagation algorithm for training to minimize the prediction error by continuously adjusting the network weights.

[0155] The training parameters include:

[0156] Learning rate: For example, set it to 0.3 to control the step size of weight update.

[0157] Momentum factor: The momentum factor can be set to 0.95 to smooth the path of gradient descent.

[0158] Iterations: Usually a higher number of iterations is set, such as 2000, to ensure that the network has sufficient learning time.

[0159] Error target: defines the threshold at which the network reaches the desired accuracy, for example, setting the error target to 0.001.

[0160] S443. Input the training samples into the neural network, and continuously adjust the weights through the back propagation algorithm to gradually reduce the output error until the set error target is reached.

[0161] S444. After the training is completed, the test samples are input into the neural network for performance testing to verify the accuracy of the network's fault classification of unseen data.

[0162] The BP neural network for general circuits also needs to normalize the fault feature vector, build a training data set, set the learning rate, momentum factor, number of iterations and error target. During the training process, the network continuously adjusts the weights through the error back propagation algorithm, and finally makes the output error reach the expected target. This process is exactly the same as the BP neural network training for specific circuits.

[0163] The network structure of the general-purpose circuit also follows the three-layer structure of input layer, hidden layer and output layer. The number of input layer nodes is consistent with the dimension of the feature vector, the number of hidden layer nodes is determined by an empirical formula, and the number of output layer nodes is set according to the number of fault modes. The number of neurons and activation function selection of the general-purpose network are also similar to those of the specific network to ensure the adaptability of the model to different fault characteristics.

[0164] Therefore, the pre-trained BP neural network for general-purpose circuits has no significant difference from the network for specific circuits in terms of training methods and structural settings, and the implementation process of specific circuits can be directly used.

[0165] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0166] In one embodiment, a BP neural network-based hemodialysis equipment analog circuit fault diagnosis device is provided, and the BP neural network-based hemodialysis equipment analog circuit fault diagnosis device corresponds to the BP neural network-based hemodialysis equipment analog circuit fault diagnosis method in the above embodiment. Figure 8 As shown, the BP neural network-based hemodialysis equipment analog circuit fault diagnosis device includes a spectrum analysis module, a fault circuit classification module, a fault feature extraction module and an analysis module. The detailed description of each functional module is as follows:

[0167] A spectrum analysis module, used to obtain input signals and process them to obtain spectrum analysis;

[0168] A fault circuit classification module is used to calculate the power spectrum density of the signal and analyze the energy distribution of the signal at different frequencies to determine whether the fault circuit is a high-frequency signal output circuit or a low-frequency signal output circuit;

[0169] A fault feature extraction module is used to analyze the frequency magnitude of the energy concentration area of ​​the high-frequency signal output circuit to determine the corresponding circuit and obtain the fault feature; wherein the corresponding circuit includes a bubble detection circuit and a heart rate monitoring circuit, and the energy concentration area is a frequency area corresponding to the energy portion exceeding a preset ratio;

[0170] If it is a low-frequency signal output circuit, a frequency threshold is set to distinguish between the high-frequency area and the low-frequency area. If the energy of the low-frequency area is higher than that of the high-frequency area, a multi-resolution analysis is performed on the input signal to obtain the fault characteristics; if the energy of the low-frequency area is not higher than that of the high-frequency area, a wavelet packet analysis is performed to obtain the fault characteristics;

[0171] The analysis module is used to analyze the fault characteristics and obtain the fault analysis results.

[0172] For the specific definition of the hemodialysis equipment analog circuit fault diagnosis device based on BP neural network, please refer to the definition of the hemodialysis equipment analog circuit fault diagnosis method based on BP neural network in the above text, which will not be repeated here. Each module in the above-mentioned hemodialysis equipment analog circuit fault diagnosis device based on BP neural network can be implemented in whole or in part by software, hardware and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0173] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig. 9 As shown. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used for data related to a method for diagnosing faults in a simulated circuit of a hemodialysis device based on a BP neural network. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for diagnosing faults in a simulated circuit of a hemodialysis device based on a BP neural network is implemented.

[0174] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the above-mentioned embodiment of the hemodialysis equipment simulation circuit fault diagnosis method based on BP neural network when executing the computer program. Alternatively, the processor implements the functions of each module / unit of the above-mentioned embodiment of the hemodialysis equipment simulation circuit fault diagnosis device based on BP neural network when executing the computer program. To avoid repetition, it will not be repeated here.

[0175] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the method for diagnosing a fault in a simulated circuit of a hemodialysis device based on a BP neural network in the above embodiment is implemented. Alternatively, when the computer program is executed by a processor, the functions of each module / unit in the device for diagnosing a fault in a simulated circuit of a hemodialysis device based on a BP neural network in the above device embodiment are implemented. To avoid repetition, it will not be described here.

[0176] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments of the present application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0177] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0178] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A method for dialysis equipment analog circuit fault diagnosis based on BP neural network, characterized in that: The following steps are involved: S1. Get the input signal and process it to get spectrum analysis; S2. Calculate the power spectrum density of the signal and analyze the energy distribution of the signal at different frequencies to determine whether the faulty circuit is a high-frequency signal output circuit or a low-frequency signal output circuit; S3. If it is a high-frequency signal output circuit, the frequency magnitude of its energy concentration area is analyzed to determine the corresponding circuit and obtain the fault characteristics; wherein the corresponding circuit includes a bubble detection circuit and a heart rate monitoring circuit, and the energy concentration area is a frequency area corresponding to the energy portion exceeding a preset ratio; If it is a low-frequency signal output circuit, a frequency threshold is set to distinguish between the high-frequency area and the low-frequency area. If the energy of the low-frequency area is higher than that of the high-frequency area, a multi-resolution analysis is performed on the input signal to obtain the fault characteristics; if the energy of the low-frequency area is not higher than that of the high-frequency area, a wavelet packet analysis is performed to obtain the fault characteristics; S4. Analyze the fault characteristics and obtain the fault analysis results; The S2 includes: S21. Calculate the power spectral density using the Welch method; S22. Calculate the total energy of the high frequency region and the low frequency region respectively according to the calculated power spectrum density curve, wherein the frequency range of the high frequency region and the frequency range of the low frequency region are preset; S23. Calculate the energy ratio of the high-frequency region and the low-frequency region. If the energy ratio is greater than a preset threshold, it is determined that the signal is mainly high-frequency energy and the fault circuit is a high-frequency signal output circuit; if the energy ratio is less than a preset threshold, it is determined that the energy is mainly low-frequency energy and the fault circuit is a low-frequency signal output circuit; The S4 includes: Inputting the fault characteristics of the low-frequency signal output circuit into the pre-trained BP neural network of the corresponding general-purpose circuit to obtain a fault analysis result, wherein the general-purpose circuit is a multi-type circuit whose total energy is concentrated in the low-frequency region; the pre-trained BP neural network of the general-purpose circuit trains the BP neural network according to a sample set of fault characteristic vectors of the low-frequency signal output circuit with pre-calibrated fault types; Inputting the fault characteristics of the high-frequency signal output circuit into the pre-trained BP neural network of the corresponding specific circuit to obtain the fault analysis result, wherein the specific circuit is a specific circuit whose total energy is concentrated at a certain frequency order of magnitude in the high-frequency region; the pre-trained BP neural network of the specific circuit is a BP neural network trained based on a sample set of fault characteristic vectors of the high-frequency signal output circuit with calibrated fault types to output the fault mode; The BP neural network for a specific circuit is trained on the high-frequency characteristics of the specific circuit; the BP neural network for a general-purpose circuit is generalized for low-frequency circuits.

2. The BP neural network-based hemodialysis equipment analog circuit fault diagnosis method according to claim 1, characterized in that: The S1 includes: S11. Select signal collection points and connect; S12. Perform signal acquisition; S13. Performing a fast Fourier transform on the collected signal to obtain a spectrum distribution; S14. Extract frequency components and corresponding amplitude information.

3. The BP neural network-based hemodialysis equipment analog circuit fault diagnosis method according to claim 2, characterized in that: The training steps of the pre-trained BP neural network of the specific circuit include: S41. Perform analog circuit simulation to determine the components that may affect the circuit, and then determine the type of fault that may occur; S42. Extracting a fault feature vector, wherein the fault mode and the fault feature vector have a clear correspondence with each other; S43. Processing the fault feature vector based on the normalization method to form the input and output sample sets required for neural network training; S44. Build and train neural networks.

4. The method for diagnosing faults of hemodialysis equipment analog circuits based on BP neural network according to claim 3, characterized in that: The training steps of the pre-trained BP neural network of the specific circuit include: S441. Setting the basic structure of the BP neural network, the basic structure includes the number of nodes in the input layer, the hidden layer and the output layer; S442. Use the error back propagation algorithm for training to minimize the prediction error by continuously adjusting the network weights; S443. Input the training samples into the neural network, and continuously adjust the weights through the back propagation algorithm to gradually reduce the output error until the set error target is reached; S444. After the training is completed, the test samples are input into the neural network for performance testing to verify the accuracy of the network's fault classification of unseen data.

5. A BP neural network-based hemodialysis equipment analog circuit fault diagnosis device, characterized in that: The method for diagnosing a fault in a hemodialysis device analog circuit based on a BP neural network according to any one of claims 1 to 4 comprises: A spectrum analysis module, used to obtain input signals and process them to obtain spectrum analysis; A fault circuit classification module is used to calculate the power spectrum density of the signal and analyze the energy distribution of the signal at different frequencies to determine whether the fault circuit is a high-frequency signal output circuit or a low-frequency signal output circuit; The fault feature extraction module analyzes the frequency magnitude of the energy concentration area of ​​the high-frequency signal output circuit to determine the corresponding circuit and obtain the fault feature; wherein the corresponding circuit includes the bubble detection circuit and the heart rate monitoring circuit, and the energy concentration area is the frequency area corresponding to the energy part exceeding the preset ratio; If it is a low-frequency signal output circuit, a frequency threshold is set to distinguish between the high-frequency area and the low-frequency area. If the energy of the low-frequency area is higher than that of the high-frequency area, a multi-resolution analysis is performed on the input signal to obtain the fault characteristics; if the energy of the low-frequency area is not higher than that of the high-frequency area, a wavelet packet analysis is performed to obtain the fault characteristics; The analysis module is used to analyze the fault characteristics and obtain the fault analysis results.

6. A computer device, characterized in that: It includes: one or more processors; Memory; One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to: execute the BP neural network-based hemodialysis equipment simulation circuit fault diagnosis method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that: The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement the BP neural network-based hemodialysis equipment simulation circuit fault diagnosis method as described in any one of claims 1 to 4.

Images