Multi-dimensional state monitoring and health management method and device for nitrification machine

Abstract

The invention discloses a nitrification machine multi-dimensional state monitoring and health management method and device, and relates to the field of nitrification machine state monitoring and health management. A full-process technical framework for multi-dimensional state monitoring and health management of the nitrification machine is provided, and equipment state monitoring parameter selection, multi-dimensional data acquisition, weak fault feature extraction, fault diagnosis, fault early warning, health state evaluation and technical verification are covered; a nitrification machine fault diagnosis model combining deep learning and an attention mechanism is provided, data and a fault mechanism are fused, and the interpretability of a fault diagnosis result is improved; a frequency spectrum combined beta-VAE model is proposed to carry out nitrification machine fault characteristic signal difference degree detection, and intelligent early warning is realized; a health state evaluation model fusing a fault mechanism is provided, and a reference is provided for health degree evaluation of fire-related equipment; predictive maintenance of the nitrification machine is achieved, work of man-machine isolation, machine replacement and black light factories is promoted, and intelligent transformation of fire-related enterprises is accelerated.

Description

Technical Field

[0001] The present invention belongs to the technical field of nitrification machine status monitoring and health management, and in particular relates to a nitrification machine multi-dimensional status monitoring and health management method and device. Background Art

[0002] The key equipment of the military fire-related production line is usually composed of many components, most of which are mechanical, electrical, gas and liquid integrated, and the relationship between different components is complex and highly coupled, which makes it difficult to monitor the equipment status and early warning of faults. The entire workflow of the military fire-related production line has the characteristics of timing and process. Any problem in any process will lead to functional failure, so the failure rate of the production process is relatively high. As one of the key equipment of the fire-related production line, the nitrifier is mainly composed of a stirring system, a separation system, a coil cooling system, an inlet and outlet pipe, a control instrument, etc., and is the core machinery for manufacturing dangerous goods. Among them, the stirring system is a component used to mix reactants, usually composed of motors, stirrers, shafts and bearings. The stirrers, shafts and other components of the nitrifier stirring system are in direct contact with the raw materials, semi-finished products or finished products of dangerous goods during the working process, and are easily affected by dynamic factors such as raw materials, equipment parameters and working environment during the working process. Once a key component fails, the shutdown will affect production efficiency at the least, and it is easy to cause product quality accidents or safety accidents at the worst, causing economic losses and casualties.

[0003] For the on-site maintenance technology of nitrifiers, fire-related enterprises often use the risk matrix method to divide the shifts to maintain nitrifiers and other equipment. The first level is regular maintenance. Regular maintenance is a maintenance method in which the equipment designer stipulates the daily inspection, weekly inspection, monthly inspection and other inspection systems of the nitrifier according to the structural characteristics of the equipment, and regularly performs lubrication, tightening, calibration, cleaning and other operations on the components or a certain combination unit of the nitrifier. Regular maintenance is a static maintenance strategy that ignores the randomness of fault occurrence. The second level is post-diagnosis, which is a diagnosis after the fault occurs. The steps are as follows: ① The maintenance personnel observe the operating status of the nitrifier to preliminarily determine whether the equipment has a fault; ② By using various measuring instruments, such as multimeters, oscilloscopes, etc., the electrical parameters, mechanical parameters and other aspects of the nitrifier are measured to determine whether the equipment has a fault; ③ By analyzing the operating status, fault phenomenon, historical records and other aspects of the equipment, the cause of the nitrifier failure is determined. Post-diagnosis is effective for some obvious fault problems, but it is difficult to diagnose complex fault types. In some cases, the nitrifier needs to be disassembled, which will affect the normal operation of the entire equipment. The third level is post-maintenance, which is repair after the fault occurs and is diagnosed. If the fault of the nitrifier involves the original manufacturer, it is necessary to contact the manufacturer to replace the relevant parts; if the fault of the nitrifier is internal, the equipment needs to be disassembled for repair. Post-maintenance is repair after the fault occurs, which leads to unstable production, large fluctuations in product quality, and is prone to safety accidents.

[0004] In summary, there is an urgent need to establish a complete technical framework for condition monitoring and health management to be applied to nitrification machines, to perform "predictive maintenance" on nitrification machines, and to ensure the safety and stability of nitrification machine operation. Summary of the invention

[0005] In order to solve the above problems, the present invention proposes a multi-dimensional state monitoring and health management method and device for a nitrification machine, establishes a complete multi-dimensional state monitoring and health management technical architecture for the nitrification machine, improves the reliability and safety of the nitrification machine operation, and lays a foundation for stable and continuous production and stable product quality.

[0006] In a first aspect, the present invention provides a multi-dimensional state monitoring and health management method for a nitrification machine, comprising:

[0007] Selecting a nitrification machine state monitoring parameter to obtain multi-dimensional data of the nitrification machine; the multi-dimensional data at least includes vibration data, current data and temperature data of the nitrification machine;

[0008] Performing fault diagnosis on the nitrification machine according to the multi-dimensional data to obtain a fault diagnosis result; and / or, performing fault warning on the nitrification machine according to the multi-dimensional data to obtain a fault warning result; and / or, performing health status assessment on the nitrification machine according to the multi-dimensional data to obtain a health status assessment result;

[0009] Display the fault diagnosis results, and / or fault warning results, and / or health status assessment results;

[0010] Wherein, the fault diagnosis of the nitrification machine is performed according to the multi-dimensional data to obtain the fault diagnosis result, including:

[0011] Performing a first preprocessing and feature extraction on the multi-dimensional data to obtain a multi-dimensional data feature set of the nitrification machine;

[0012] Inputting the multi-dimensional data feature set of the nitrification machine into the trained nitrification machine fault diagnosis model to obtain a fault category label;

[0013] Inputting the fault category label into a pre-built fault mode library to obtain a fault diagnosis result of the nitrification machine;

[0014] The method of performing fault warning on the nitrification machine according to the multi-dimensional data to obtain a fault warning result includes:

[0015] Performing a second preprocessing on the multi-dimensional data to obtain spectrum data;

[0016] Inputting the spectrum data into a pre-built nitrification machine fault warning model to obtain a fault warning result;

[0017] The health status evaluation of the nitrification machine is performed according to the multi-dimensional data to obtain a health status evaluation result, including:

[0018] The multi-dimensional data is input into a nitrification machine health status assessment model pre-constructed based on a nitrification machine index system, and a health status score is performed on the nitrification machine to obtain a health status assessment result.

[0019] In an optional implementation, the trained nitrification machine fault diagnosis model is obtained by the following method:

[0020] Obtain fault diagnosis training sample data;

[0021] Performing a first preprocessing on the fault diagnosis training sample data to obtain the first preprocessed fault diagnosis training sample data;

[0022] Performing feature extraction on the fault diagnosis training sample data after the first preprocessing to obtain a nitrification machine training feature set;

[0023] The nitrification machine training feature set is input into a pre-built nitrification machine fault diagnosis model for training to obtain a trained nitrification machine fault diagnosis model; wherein the nitrification machine fault diagnosis model includes an improved gated recurrent unit network module, a one-dimensional convolutional neural network module, an attention mechanism module and a Softmax classifier module, the improved gated recurrent unit network module and the one-dimensional convolutional neural network module are respectively connected to the Softmax classifier module, and the attention mechanism module is connected to the improved gated recurrent unit network module;

[0024] The improved gated recurrent unit network module is obtained by the following method:

[0025] Construct a gated recurrent unit network module;

[0026] Using a one-dimensional global pooling layer to replace the output layer of the gated recurrent unit network module to obtain an improved gated recurrent unit network primary module;

[0027] The attention mechanism module is used to adjust the output layer of the improved gated recurrent unit network primary module to obtain an improved gated recurrent unit network module.

[0028] In an optional embodiment, the fault diagnosis training sample data includes first fault diagnosis training sample data and second fault diagnosis training sample data; wherein, the first fault diagnosis training sample data is obtained by performing fault simulation on a pre-built nitrification machine fault simulation test bench; the second fault diagnosis training sample data is constructed using a pre-built interference model and normal samples; the normal samples are obtained by simulating the normal working state of the nitrification machine on the nitrification machine fault simulation test bench.

[0029] In an optional embodiment, the feature extraction is performed on the first preprocessed fault diagnosis training sample data to obtain a nitrification machine training feature set, including:

[0030] A weak signal is extracted from the fault diagnosis training sample data after the first preprocessing by using the ARLW algorithm, and the weak signal is enhanced by using the SPM algorithm to obtain an enhanced weak signal; the enhanced weak signal is impact quantized by using the Hilbert envelope demodulation algorithm and the peak energy value to obtain weak signal feature data; and multidimensional feature data is extracted from the fault diagnosis training sample data after the first preprocessing; wherein the multidimensional feature data includes time domain features, frequency domain features and time-frequency domain features;

[0031] The weak signal feature data and the multi-dimensional feature data are used to construct the nitrification machine training feature set.

[0032] In an optional embodiment, the nitrification machine fault warning model includes a beta-variational autoencoder module, an error distribution module and a comparison module connected in sequence;

[0033] The beta-variational autoencoder module is used to perform spectrum reconstruction on the spectrum data using a trained beta-variational autoencoder model to obtain spectrum reconstructed data;

[0034] The error distribution module is used to calculate the error distribution of the spectrum reconstruction data to obtain first error distribution data;

[0035] The comparison module is used to compare the first error distribution data with a predetermined first warning threshold, and use the comparison result as a fault warning result.

[0036] In an optional embodiment, the trained beta-variational autoencoder model is determined by the following method:

[0037] Acquire fault warning training sample data; wherein the fault warning training sample data is obtained by simulating the normal working state of the nitrifier through a pre-built nitrifier fault simulation test bench;

[0038] Performing a second preprocessing on the fault warning training sample data to obtain spectrum training data; wherein the second preprocessing includes Fourier transform;

[0039] The spectrum training data is input into a pre-built beta-variational autoencoder model for training to obtain a trained beta-variational autoencoder model and spectrum reconstruction training data.

[0040] In an optional embodiment, the method further comprises:

[0041] The multi-dimensional data is input into a nitrifying machine health status assessment model pre-constructed based on a nitrifying machine index system, and a health status score is performed on the nitrifying machine to obtain a health status assessment result; wherein, the index system is constructed based on the status and structural composition of the nitrifying machine; the status includes the operating condition, fault and maintenance condition, and environmental condition of the nitrifying machine; and the structural composition includes a shaft, bearings, a motor, and an agitator.

[0042] In an optional embodiment, the operating conditions include an operating time score and a degradation degree; the fault and maintenance conditions include a mean time between failures score; the environmental conditions include an environmental temperature score and an environmental humidity score;

[0043] The running time score is determined by the following formula (1):

[0044]

[0045] In formula (1), YE represents the operating time score; x1 represents the number of years that the key equipment of the nitrification machine has been put into use; k1 represents the design life of the key equipment of the nitrification machine;

[0046] The degradation degree is determined by the following equations (2) and (3):

[0047]

[0048] In formula (2) and formula (3), d j Indicates the degree of degradation; x i represents the actual measured value of the characteristic parameters of the key equipment of the nitrification machine; f Indicates the normal value or factory value of the characteristic parameters of the key equipment of the nitrification machine; x max Indicates the upper limit characteristic parameter threshold when the key equipment of the nitrification machine needs to be shut down; x min represents the lower limit characteristic parameter threshold when the key equipment of the nitrification machine is shut down; k2 represents x i The relationship between the health status of the key equipment of the nitrification machine and the i It is linearly related to the nitrification machine or key equipment of the nitrification machine;

[0049] When the characteristic parameter threshold x max and x min When is a range threshold, the degradation degree is determined by the following formula (4):

[0050]

[0051] In formula (4), x1 and x2 respectively represent the lower limit and upper limit of the characteristic parameters of the key equipment of the nitrification machine when the parameters are normal, and x0 and x3 respectively represent the lower limit and upper limit of the characteristic parameters when the key equipment of the nitrification machine must be shut down;

[0052] The mean time between failures score is determined by the following formula (5):

[0053]

[0054] In formula (5), M1 represents the mean time between failures score; MTBF represents the mean time between failures, and the mean time between failures is determined by the ratio of the total operating time of the nitrifier to the total number of failures;

[0055] The ambient temperature score is determined by the following formula (6):

[0056]

[0057] In formula (6), T represents the ambient temperature score, t2 represents the current ambient temperature; the normal range threshold of ambient temperature is defined as X4℃~Y4℃; When T=100;

[0058] The environmental humidity score is determined by the following formula (7):

[0059]

[0060] In formula (7), H represents the environmental humidity score, t3 represents the current environmental humidity; the normal range threshold of environmental absolute humidity is defined as M2g / m 3 ~N2g / m 3 Among them, when , H=100.

[0061] In an optional implementation, the fault diagnosis result includes fault characteristics and fault causes.

[0062] In a second aspect, the present invention provides a multi-dimensional state monitoring and health management device for a nitrification machine, comprising:

[0063] A data acquisition module, used to acquire multi-dimensional data of the nitrification machine; the multi-dimensional data at least includes vibration data, current data and temperature data of the nitrification machine;

[0064] an intelligent diagnosis and early warning module, for performing fault diagnosis on the nitrification machine according to the multi-dimensional data to obtain a fault diagnosis result; and / or performing fault early warning on the nitrification machine according to the multi-dimensional data to obtain a fault early warning result; and / or performing health status assessment on the nitrification machine according to the multi-dimensional data to obtain a health status assessment result;

[0065] A display module, used to display the fault diagnosis result, and / or the fault warning result, and / or the health status assessment result;

[0066] Wherein, the intelligent diagnosis and early warning module includes a fault diagnosis module, and / or a fault early warning module, and / or a health status assessment module;

[0067] The fault diagnosis module comprises:

[0068] A first processing module, used for performing a first preprocessing and feature extraction on the multi-dimensional data to obtain a multi-dimensional data feature set of the nitrification machine;

[0069] A diagnosis module, used for inputting the multi-dimensional data feature set of the nitrification machine into a trained nitrification machine fault diagnosis model to obtain a fault category label;

[0070] A first result module is used to input the fault category label into a pre-built fault mode library to obtain a fault diagnosis result of the nitrification machine;

[0071] The fault warning module comprises:

[0072] A second processing module, used for performing a second preprocessing on the multi-dimensional data to obtain spectrum data;

[0073] The second result module is used to input the spectrum data into a pre-built nitrification machine fault warning model to obtain a fault warning result;

[0074] The health status assessment module is used to input the multi-dimensional data into a nitrification machine health status assessment model pre-constructed based on a nitrification machine index system, perform a health status score on the nitrification machine, and obtain a health status assessment result.

[0075] The beneficial effects of the technical solution provided by the embodiment of the present invention are as follows: the multi-dimensional state monitoring and health management method and device of the nitrifier of the present invention provide a technical framework for fault diagnosis, fault warning and health status assessment; fault diagnosis is: by performing a first preprocessing and feature extraction on the multi-dimensional data of the nitrifier to obtain a multi-dimensional data feature set of the nitrifier, and then inputting the obtained multi-dimensional data feature set of the nitrifier into a trained nitrifier fault diagnosis model to obtain a fault category label, and finally inputting the fault category label into a pre-built fault mode library to obtain a fault diagnosis result of the nitrifier; fault warning is: performing a second preprocessing on the multi-dimensional data of the nitrifier .... Processing is performed to obtain spectrum data; the spectrum data is then input into a pre-constructed nitrification machine fault warning model to obtain a fault warning result; the present invention also scores the health status of the nitrification machine according to the multi-dimensional data and a nitrification machine health status assessment model pre-constructed based on the nitrification machine indicator system to obtain a health status assessment result; the present invention can perform fault diagnosis and fault warning on the nitrification machine through the established nitrification machine fault diagnosis, fault warning and health status assessment technical architecture, realize "predictive maintenance" of the nitrification machine, and improve the reliability, stability and safety of the nitrification machine operation, thereby laying a foundation for stable and continuous production and stable product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 A flow chart of a multi-dimensional state monitoring and health management method for a nitrification machine provided in an embodiment of the present invention;

[0077] Figure 2 The embodiment of the present invention provides Figure 1 Schematic diagram of the specific method flow of step S210;

[0078] Figure 3 The embodiment of the present invention provides Figure 1 Schematic diagram of the specific method flow of step S220;

[0079] Figure 4 A schematic diagram of the technical framework of the entire process of a nitrification machine provided by an embodiment of the present invention;

[0080] Figure 5 A schematic diagram of the frequency response of the ARLW provided in an embodiment of the present invention;

[0081] Figure 6 A schematic diagram of a fault diagnosis process of a nitrification machine driven by both mechanism and data according to an embodiment of the present invention;

[0082] Figure 7 A schematic diagram of a data-based nitrification machine fault diagnosis model algorithm flow chart provided in an embodiment of the present invention;

[0083] Figure 8 A schematic diagram of the change of reconstruction error over time provided by an embodiment of the present invention;

[0084] Fig. 9 A schematic diagram of a health status assessment process for key equipment of a nitrification machine provided in an embodiment of the present invention;

[0085] Fig.10 This is a schematic diagram of a multi-dimensional state monitoring and health management device system for a nitrification machine provided in an embodiment of the present invention.

[0086] In the figure: 100-data acquisition module; 200-intelligent diagnosis and early warning module; 210-fault diagnosis module; 211-first processing module; 212-diagnosis module; 213-first result module; 220-fault early warning module; 221-second processing module; 222-second result module; 230-health status assessment module; 300-display module. DETAILED DESCRIPTION

[0087] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0088] At present, in terms of data collection of nitrifiers, the fire-related enterprises have the sensing monitoring including material flow, temperature, liquid level, current, etc., as well as relevant process parameters reflecting material feeding and interlocking alarm signals. Although real-time sensing monitoring of equipment temperature, vibration, pressure and other health conditions has been carried out, fault diagnosis and fault warning of nitrifiers have not been achieved.

[0089] In addition, during the actual operation of the nitrifier, fault signals are often difficult to collect, resulting in a small number of fault samples, but the consequences of the fault are very serious. Therefore, it is urgent to expand the fault samples by building a fault simulation test bench and adding noise to normal sample data to achieve data enhancement and fault diagnosis and early warning of the nitrifier.

[0090] In the fault diagnosis of nitrifiers, the technical methods currently used are model-driven, knowledge-driven and data-driven. The model-driven method accurately constructs the mechanism model of the equipment based on theories such as mathematics and dynamics; the knowledge-driven method encodes the knowledge and experience of experts into rules or decision trees. The data-driven method can be divided into combining signal processing with machine learning and directly using deep learning methods to achieve end-to-end fault diagnosis. Among them, signal processing methods include empirical mode decomposition, wavelet transformation, variational mode decomposition and other methods. These methods are used to extract time domain, frequency domain and time-frequency domain features in the data, and then support vector machine, K nearest neighbor, artificial neural network and other machine learning methods are used to classify and learn the features to achieve fault diagnosis. With the increasing maturity of artificial intelligence technology, deep learning methods are used to identify equipment faults. This method can extract features layer by layer and solve the limitations of traditional machine learning methods. However, the single model fault diagnosis accuracy is limited, and the deep learning model has a "black box" problem. Moreover, the more parameters the model has, the greater the amount of information stored, which will bring about the problem of information overload and the problem of poor interpretability of the diagnosis results. Therefore, it is difficult to directly apply it to the fault diagnosis and operation and maintenance of fire-related equipment.

[0091] In terms of the health status assessment of nitrifiers, the technical methods used include model-driven, knowledge-driven and data-driven. The model-driven method mainly constructs a dynamic model of the equipment to accurately identify the early degradation points of the equipment. The knowledge-driven health status assessment process is to establish a comprehensive assessment system for the health status of the equipment by considering many uncertain random factors such as the quality of the equipment itself, the maintenance status and the operating environment, and to use entropy weight method, analytic hierarchy process (AHP), fuzzy comprehensive evaluation and other methods to assess the health status of the equipment. The data-driven health status assessment process is: first, obtain multi-dimensional data of the equipment from sensors such as vibration and temperature; then, based on the data that can characterize the health status of the equipment, use signal processing methods, artificial intelligence algorithms, etc. to construct health indicators to represent the health status of the equipment; finally, by analyzing the degradation trend of the health indicators, a knowledge base of health status assessment criteria is constructed, and the operation process of the equipment throughout its life cycle is divided into multiple different health stages to achieve health assessment of the equipment. However, data-driven health status assessment methods rely on the completeness of the data set. However, the lack of fire-related equipment failure samples makes it difficult to use data-driven methods effectively; model-driven methods need to accurately construct the dynamic model of the observed object based on mathematics, dynamics and other theories. As equipment becomes increasingly complex and intelligent, there is a coupling mechanism between the failures of multiple mechanisms of equipment, which makes it difficult to build an accurate dynamic model. The above factors have led to little research on the health assessment system of fire-related equipment.

[0092] For the specific equipment of nitrification machine, a complete multi-dimensional state monitoring and health management technical framework has not yet been established. This framework should cover the whole process from state monitoring parameter selection, multi-dimensional data collection, weak fault feature extraction, to fault diagnosis, fault warning, health status assessment, and technical verification. This lack limits the systematicness and effectiveness of nitrification machine state monitoring and fault diagnosis, so as to improve the reliability and safety of equipment operation.

[0093] Based on this, this embodiment proposes a multi-dimensional state monitoring and health management method and device for a nitrification machine, and establishes a complete technical architecture for multi-dimensional state monitoring and health management, from state monitoring parameter selection, multi-dimensional data collection, weak fault feature extraction, to fault diagnosis, fault warning, health status assessment, and the entire process of technical verification, so as to improve the reliability and safety of equipment operation.

[0094] See also Figure 1 The present embodiment provides a multi-dimensional state monitoring and health management method for a nitrification machine, including steps S100 to S300.

[0095] Step S100, selecting a nitrification machine status monitoring parameter, based on which, obtaining multi-dimensional data of the nitrification machine; the multi-dimensional data at least includes vibration data, current data and temperature data of the nitrification machine.

[0096] This embodiment obtains the key influencing factors affecting the health of the nitrifying machine by analyzing the process flow of the nitrifying machine for producing dangerous goods and studying the mechanical structure of the nitrifying machine. According to these influencing factors, a parameter selection system for equipment perception monitoring is established, and sensors such as vibration acceleration, temperature and current are installed to realize the real-time collection of the state monitoring parameters of the nitrifying machine. The nitrifying machine state monitoring parameters finally selected include vibration parameters (for example, the impact fault vibration signal of the rolling bearing monitored by the acceleration sensor, the vibration signal monitored by the vibration sensor, etc.), temperature parameters and current parameters. Through these monitoring indicators, the abnormal signs before the fault occurs are reflected, and the accurate monitoring of the nitrifying machine and the timely discovery of the signs of accidents are realized, so that a reasonable maintenance strategy can be formulated before the accident occurs.

[0097] Specifically, the vibration data may be the vibration data of the rolling bearing, the current data refers to the current data of the nitrification machine motor, which may be obtained by a current sensor; the temperature data refers to the temperature of each component of the nitrification machine, which may be obtained by a temperature sensor.

[0098] Step S200, using multi-dimensional data to sequentially perform fault diagnosis, fault warning and health status assessment on the nitrification machine. The step of fault diagnosis is shown in step S210, the step of fault warning is shown in step S220, and the step of health status assessment is shown in step S230.

[0099] Step S210, performing fault diagnosis on the nitrification machine according to the multi-dimensional data to obtain a fault diagnosis result, specifically including the following steps S211 to S213.

[0100] See also Figure 2 , step S211, performing a first preprocessing and feature extraction on the multi-dimensional data to obtain a multi-dimensional data feature set of the nitrification machine.

[0101] Specifically, the first preprocessing can be data deduplication, data missing processing and outlier processing, and feature extraction includes feature extraction of weak signals and multi-dimensional feature extraction of multi-dimensional data. In actual scenarios, the process of a mechanical power system from a normal state to an abnormality (or failure) occurs gradually, and the abnormality that occurs is usually an early weak abnormality. When an early abnormality occurs in a nitrifier, the fault characteristics in the signal are extremely weak, and the signs are not obvious. At the same time, the working conditions are changeable and there are interference sources. And due to the limitations of installation conditions, it is sometimes difficult to measure the signal directly, and it can only be measured indirectly through the external surface of the equipment, which further causes the vibration signal generated by the excitation source to attenuate. Based on the above characteristics, there is a problem in the signal monitoring process that weak fault characteristics are submerged in large vibration signals. Therefore, it is necessary to carry out research on the feature extraction technology of weak signals of the whole machine in the early stage of vibration faults to solve the problem of extracting useful information from the early signals of nitrifier faults. This embodiment uses an antisymmetric real Laplace wavelet (ARLW) algorithm and a spatial pyramid matching (SPM) algorithm to extract weak signals and enhance characteristic signals.

[0102] Step S212, inputting the multi-dimensional data feature set of the nitrification machine into the trained nitrification machine fault diagnosis model to obtain a fault category label.

[0103] Step S213, input the fault category label into the pre-constructed fault mode library to obtain the fault diagnosis result of the nitrifier. Construct a complete fault chain model including the fault cause, fault path and fault mode, establish the fault diagnosis rules of the nitrifier, and obtain the fault mode library. The fault mode library can quickly trace the fault when the nitrifier fails. This embodiment combines expert knowledge and fault tree analysis method to construct a complete fault chain model of the nitrifier including the fault cause, fault path and fault mode. Based on the fault chain model, the fault rules are extracted. When the nitrifier fails, fault reasoning is carried out according to the fault rules to quickly infer the cause of the current fault, thereby realizing rapid tracing of the fault. This embodiment solves the problem of "black box" and poor interpretability of diagnostic results in neural networks by constructing a fault mode library, which is conducive to application in fault diagnosis and operation and maintenance.

[0104] In step S212, the trained nitrification machine fault diagnosis model is obtained through the following steps (1) to (3).

[0105] Step (1), obtaining fault diagnosis training sample data; wherein the fault diagnosis training sample data is multi-dimensional data about the nitrification machine, and the multi-dimensional data includes vibration parameters, temperature parameters and current parameters.

[0106] Here, the fault diagnosis training sample data includes first fault diagnosis training sample data and second fault diagnosis training sample data. The first fault diagnosis training sample data is obtained by performing fault simulation on a pre-built nitrification machine fault simulation test bench.

[0107] Specifically, the fault simulation test bench includes a platform base plate, a motor, a torque and speed sensor, a rolling bearing rotor system, a coupling, an agitator, an electrical control system, and a monitoring and diagnosis analysis system. These components are used to simulate the interaction of various components in the operation of the nitrifier, and the transmission and influence of faults between various components. The fault simulation test bench is also equipped with the required typical rolling bearing fault kit, agitator fault kit, and motor fault simulation kit. Fault simulation is achieved by prefabricating typical fault defects, including motor stator windings, rotor broken bars, built-in fault bearings, agitator impeller mass imbalance, rolling bearing inner ring pitting, outer ring pitting, rolling element pitting, etc. The monitoring and diagnosis analysis system mainly includes sensors, acquisition system hardware, and monitoring and analysis software. The acquisition hardware includes acceleration sensors and supporting components, speed sensors and supporting components, temperature sensors and supporting components, and current sensors and supporting components. The vibration acceleration sensor is placed on the rolling bearing seat; the speed measurement uses a photoelectric sensor, the sensor can be fixed to the base using a magnetic table seat, and the reflective sheet is attached to the surface of the shaft journal; the temperature sensor is used to measure the temperature of each component; the current sensor is used to measure the motor current. The acquisition system hardware includes acceleration acquisition module, key phase acquisition module, main control module, etc.; the monitoring system software mainly includes data acquisition module, data management module, data processing module and fault analysis and diagnosis module. Among them, the key phase acquisition module is mainly used to monitor the speed on the variable frequency drive equipment of the nitrification machine. When the equipment starts or stops, it must pass the critical speed (except for the rigid shaft). This embodiment uses the key phase to match the shaft vibration probe, which can perfectly capture the vibration trend when starting / stopping.

[0108] According to the mechanical structure and operation principle of the nitrification machine, this embodiment decomposes it into motor, shaft, bearing, and agitator structures, and then builds a nitrification machine fault simulation test bench based on these structures. The nitrification machine fault simulation test bench simulates the data changes under various fault scenarios and generates diversified fault sample data. This embodiment simulates and obtains the fault mechanism through the fault simulation test bench to realize the data enhancement technology research on the nitrification machine equipment.

[0109] For the second fault diagnosis training sample data, firstly, based on the fault mode analysis of the nitrifier and the data distribution of the fault samples, a reasonable interference model is designed, including adding noise and changing parameter trends; then the interference is applied to the normal samples to simulate the data changes under various fault scenarios and generate diversified fault sample data. The normal samples here are obtained by simulating the normal working state of the nitrifier on the nitrifier fault simulation test bench.

[0110] This embodiment obtains fault samples by building a fault simulation test bench and fault simulation, conducts model training and studies the fault mechanism, and further enriches the number of samples by designing a reasonable interference model and applying it to normal samples. The above two methods solve the problem of insufficient data volume from the root of the data.

[0111] Step (2), extracting features from the fault diagnosis training sample data to obtain a nitrification machine training feature set, specifically includes the following steps (21) to (23).

[0112] Step (21), using the ARLW algorithm to extract the weak signal in the fault diagnosis training sample data, using the SPM algorithm to enhance the weak signal to obtain an enhanced weak signal; using the Hilbert envelope demodulation algorithm and the peak energy value to perform impact quantization on the enhanced weak signal to obtain weak signal feature data.

[0113] Specifically, the fault damage signal becomes a weak signal after being transmitted through a complex path, but the characteristics do not completely disappear. The ARLW algorithm and the SPM algorithm are used to extract the fault impact signal and enhance the characteristic signal, and the weak signal characteristics are extracted, including the following steps (a) to (c).

[0114] Step (a) Weak signal extraction based on ARLW algorithm.

[0115] Firstly, the weak signal in the fault diagnosis training sample data is extracted by using the ARLW algorithm. The impact fault vibration signal of the rolling bearing can be simplified into a single-cycle impact mechanical vibration fault signal model of an approximate linear system, as shown in the following equation (a-1).

[0116]

[0117] In formula (a-1), m is the mass of the vibration system, ξ is the system damping ratio, ω d The system has a damped natural frequency.

[0118] Based on the above model and the idea of ​​sparse signal expression, a wavelet filter, antisymmetric real Laplace wavelet (ARLW), is constructed directly from the impact fault vibration signal model, as shown in equation (a-2).

[0119]

[0120] This wavelet function is similar to the real part of the complex Laplace wavelet, but with a phase delay and an antisymmetric extension. Figure 5 The frequency response curve of the ARLW is shown, showing that it is a bandpass filter with zero phase characteristics, which is crucial for signal reconstruction and extraction.

[0121] In summary, ARLW is regarded as the unit impulse response of a bandpass filter, so bandpass filtering can be achieved as shown in the following formula (a-3).

[0122] WT x(t) (ξ,ω d )=F -1 {X(f)Ψ * (f)}, (a-3);

[0123] In formula (a-3), F -1 represents the inverse Fourier transform operator, x(t) is the input vibration signal, x(f) is the Fourier transform of the signal x(t), and f is the frequency variable (unit: Hz).

[0124] The calculation of the half-power bandwidth of the ARLW wavelet filter is shown in formula (a-4).

[0125]

[0126] In formula (a-4), Let f d =ω d / 2π, then ψ(t) is as shown in formula (a-5).

[0127]

[0128] (b) Feature signal enhancement based on SPM algorithm.

[0129] The characteristic signal enhancement based on the SPM algorithm simulates the physical sensor resonance phenomenon by constructing a digital simulation sensor, uses the resonance frequency band of the sensor frequency response function to perform pulse enhancement reconstruction on the fault characteristic frequency band, and uses sparse filtering based on a redundant dictionary to perform noise reduction, and finally obtains the enhanced fault signal; specifically, it includes the following steps ① to ③.

[0130] Step ①, construct a digital analog sensor.

[0131] The amplitude-frequency and phase-frequency characteristic curve equations of an ideal vibration acceleration sensor are shown in equations (b-1) and (b-2).

[0132]

[0133] In formula (b-1) and (b-2), ω is the vibration frequency, ω n is the sensor resonant frequency, ζ is the attenuation coefficient, A(ω) is the sensor amplification factor, is the phase response.

[0134] According to the ARLW weak signal feature extraction technology, the center frequency of the optimal filtering frequency band is selected as the resonant frequency ω of the ideal sensor n =f c-opt The nonlinear least squares method is used to fit the frequency response curve measured by the actual sensor, and the frequency points and their corresponding magnifications are used to determine the unknown parameters ζ and ω in the sensor frequency response function. n , so that the standard deviation of the function is minimized.

[0135] Step ②: Reconstruction of fault signal impact enhancement.

[0136] The reconstruction process is the process of collecting signals by digital analog sensors, and the bearing fault impact signal is convolved with the ideal sensor function. First, the constructed ideal sensor function is converted into a complex form, as shown in formula (b-3).

[0137]

[0138] Then, an inverse Fourier transform is performed, and its expression is shown in (b-4).

[0139]

[0140] In formulas (b-3) and (b-4), t represents the time corresponding to the sampling point.

[0141] Finally, the time domain signal of the ideal sensor function is circularly convolved with the impact signal of the bearing fault to digitize the process of the analog sensor collecting signals and generating resonance, and finally the reconstructed signal x after impact enhancement is obtained.re (n), the calculation formula is shown in (b-5).

[0142]

[0143] Step ③: Sparse filtering denoising method based on redundant dictionary.

[0144] In the process of constructing the digital analog sensor reconstructed signal, while the fault impact is amplified, the noise signal will also be amplified synchronously, so the reconstructed signal is denoised. The sparse filtering denoising method based on redundant dictionary is an effective signal processing technology, which is mainly used to remove noise in the signal while retaining the important features of the signal. This method uses redundant dictionary and sparse representation theory to improve the denoising effect by optimizing the dictionary and sparse decomposition. This embodiment selects five signals, namely sym8 wavelet, sym8 wavelet packet, discrete cosine transform, sine signal and cosine signal, to form a redundant dictionary, sets the coefficient of weak matching pursuit to 0.6, and realizes denoising of the reconstructed signal.

[0145] Step (c), impact quantification features.

[0146] The Hibert envelope demodulation is performed on the impulse enhanced reconstructed signal after matching pursuit denoising to obtain the envelope spectrum signal. The peak energy value at the fault characteristic frequency is calculated and standardized as a characteristic indicator to measure the degree of fault. The specific calculation process is as follows.

[0147] Use the fault characteristic frequency f in the envelope spectrum d The sum of the amplitudes at its harmonics is used as the representation of the peak energy value, as shown in (c-1).

[0148]

[0149] In formula (c-1), is the envelope spectrum fault characteristic frequency f d and the sum of the peak energy values ​​at its M-order harmonics, X gSE (m·f d ) represents the signal envelope spectrum m·f d The amplitude corresponding to the frequency.

[0150] The amplitude corresponding to the amplitude of the transfer frequency component in the original signal spectrum is used as a measure of the speed and load, thereby standardizing the envelope spectrum peak energy, and then converting the unit into dB value. Finally, the average of the energy peaks at various fault characteristic frequencies and their harmonics is calculated as the final envelope spectrum energy value, as shown in (c-2).

[0151]

[0152] In formula (c-2), STX gSE (f) represents the normalized energy value at each frequency in the signal envelope spectrum, unit: dB, H(f r ) represents the amplitude corresponding to the frequency in the original signal spectrum, f r Indicates frequency conversion.

[0153]

[0154] In formula (c-3), Indicates the fault characteristic frequency f d The envelope spectrum eigenvalue of is the weak signal eigenvalue finally extracted.

[0155] Taking rolling bearings as an example, the vibration acceleration sensor is installed on the outside of the bearing seat in the vertical direction to collect signals, and the envelope spectrum eigenvalues ​​are extracted using the above algorithm process. The results show that as the fault severity gradually increases, the envelope spectrum eigenvalues ​​of the characteristic frequency of the bearing outer ring also gradually increase, and the envelope spectrum eigenvalues ​​of the characteristic frequency of bearing signals of different fault types have obvious differences.

[0156] Step (22), extracting multidimensional feature data from the fault diagnosis training sample data; wherein the multidimensional feature data includes time domain features, frequency domain features and time-frequency domain features.

[0157] Specifically, the extracted time domain features include dimensional parameters and dimensionless parameters. Dimensional parameters include peak value, peak-to-peak value, mean value, effective value, variance and other features, while dimensionless parameters include kurtosis, skewness, waveform, pulse, margin and other features. The extracted frequency domain features include center frequency, frequency standard deviation, frequency domain maximum energy and other features; time-frequency domain features include wavelet transform, empirical mode decomposition and other features. These features constitute multidimensional feature data.

[0158] Step (23), using the weak signal feature data and the multi-dimensional feature data to construct a nitrification machine training feature set.

[0159] The multi-dimensional feature data is combined with the weak signal feature data to construct a high-dimensional feature set of typical faults, which is used as the training feature set of the nitrifier.

[0160] Step (3), inputting the nitrification machine training feature set into a pre-built nitrification machine fault diagnosis model for training, and obtaining a trained nitrification machine fault diagnosis model; wherein the nitrification machine fault diagnosis model includes an improved gated recurrent unit network module (i.e., GGRU module), a one-dimensional convolutional neural network module (i.e., 1DCNN module), an attention mechanism module, and a Softmax classifier module, the improved gated recurrent unit network module and the one-dimensional convolutional neural network module are respectively connected to the Softmax classifier module, and the attention mechanism module is connected to the improved gated recurrent unit network module.

[0161] Among them, the improved gated recurrent unit network module is obtained through the following steps (31) to (33).

[0162] Step (31), construct a gated recurrent unit network module (i.e., GRU module).

[0163] Step (32), use a one-dimensional global pooling layer to replace the output layer (i.e., the flattening layer) of the gated recurrent unit network module to obtain an improved gated recurrent unit network primary module.

[0164] Step (33), use the attention mechanism module to adjust the output layer of the improved gated recurrent unit network primary module to obtain an improved gated recurrent unit network module (i.e., GGRU module).

[0165] Specifically, firstly, the one-dimensional global average pooling (1D-GAP) layer is used to replace the terminal layer of the traditional gated recurrent unit network module (GRU module) to improve the primary module of the gated recurrent unit network; this not only aggregates the influencing components of the defects in the vibration signal, but also compresses the number of model parameters and converts the data dimension. Secondly, the soft attention mechanism is applied in the output layer of the improved primary module of the gated recurrent unit network to extract key fault features from multi-dimensional data and better capture the key information in the time series data. Then, by combining the one-dimensional convolutional neural network module, the fusion of different types of frequency and sequence features is realized, solving the problem of insufficient feature extraction capability of a single network. Finally, the features output by the improved gated recurrent unit network module and the features output by the one-dimensional convolutional neural network module are fused, and the fused features are input into the Softmax classifier to achieve accurate identification of different fault locations and fault types of various components of the nitrifier.

[0166] The training principle of the nitrification machine fault diagnosis model is as follows: Figure 7 As shown, first input the data, where the data refers to the aforementioned data set. Data standardization refers to the first preprocessing of the data set. Divide the data set, that is, divide the data set into a training set, a validation set, and a test set in a ratio of 7:2:1. Input the training set and the validation set into the constructed nitrifier fault diagnosis model (i.e. Figure 7The GGRU-Attention and 1DCNN models in the nitrifier fault diagnosis model are then set in sequence (network parameters, training parameters, etc. of the nitrifier fault diagnosis model), the model is trained using the Adam optimizer and the cross entropy loss function, and the model weight parameters are updated using back propagation. After the training and verification of the training set and the verification set are completed, the trained GGRU-Attention and 1DCNN models are obtained, and the test set is used to test the trained GGRU-Attention and 1DCNN models. First, input the test set into the GGRU-Attention and 1DCNN models, extract the Attention layer features from the GGRU-Attention model, extract the FC layer features from the 1DCNN model, perform feature fusion, and then determine whether the current set is a test set. If so, execute the following steps: input the data into the trained Softmax classifier, and output the fault diagnosis result: accuracy. If not, execute the following steps: set hyperparameters, train the model using the Adam optimizer and the cross entropy loss function, and update the model weight parameters using back propagation. Determine whether the current model converges on the validation set. If not, return to the step of setting hyperparameters and continue. If yes, also execute the following steps: input the data into the trained Softmax classifier, and output the fault diagnosis result: accuracy.

[0167] like Figure 6 As shown, the aforementioned step S211 obtains the fault category label through data preprocessing (first preprocessing), data set division (divided into training set, verification set and test set), feature extraction (weak signal feature extraction and multi-dimensional feature extraction), feature fusion, model construction, and data-based fault diagnosis; the obtained fault category label is input into the pre-built fault mode library for mechanism-based fault diagnosis, and detailed information output including fault characteristics and fault causes (nitrification machine fault diagnosis result) is obtained. Among them, the process of building the fault mode library is: fault chain construction and storage, diagnosis rule establishment, and fault mode library construction and storage.

[0168] Taking the electronic rotor broken bar fault as an example, the motor rotor broken bar fault mechanism is shown in Table 1.

[0169] Table 1 Motor rotor broken bar fault mechanism

[0170]

[0171] Through the above data-driven method, the nitrification machine fault was diagnosed as a motor rotor bar failure. According to the fault mechanism in Table 1, it can be inferred that the possible causes of the motor failure are broken rotor conductors, uneven rotor mass distribution, or falling off of rotor parts. The fault characteristic is that the vibration value increases with the increase of speed and is unrelated to the motor load. Based on this, the motor is repaired accordingly.

[0172] It should be noted that the mechanism-based fault diagnosis builds a fault chain, and the data-based fault diagnosis outputs a fault category, which is a label. The label is the fault category of the nitrifier, not a one-dimensional feature data. Corresponding to the fault mode library built based on the mechanism, the fault characteristics corresponding to the label and the path of the fault can be output, as described in Table 1.

[0173] The fault diagnosis method based on dual drive of mechanism and data in this embodiment, firstly, constructs a complete fault chain model including fault cause, fault occurrence path and fault mode, establishes fault diagnosis rules for nitrifier, and realizes rapid tracing of fault when fault occurs in nitrifier. Secondly, collects field fault data and test fault simulation data, and pre-processes the collected vibration, temperature and other data; extracts time domain, frequency domain, time-frequency domain and other features, combines weak fault features, constructs a high-dimensional feature set of typical faults, and adopts feature fusion method to obtain features that can accurately characterize nitrifier faults; develops an algorithm model combining one-dimensional convolutional neural network (1DCNN), improved gated recurrent unit network (GGRU) and attention mechanism (Attention), and realizes data-driven fault diagnosis. Finally, the fault category obtained by data-based diagnosis is matched to the fault mode library constructed based on mechanism, and detailed information including fault features and fault causes of nitrifier is obtained, so as to realize intelligent diagnosis of nitrifier fault driven by both data and mechanism.

[0174] See also Figure 3 , step S220, performs fault warning on the nitrification machine according to the multi-dimensional data to obtain a fault warning result, which specifically includes the following steps S221 to S222.

[0175] Step S221, performing second preprocessing on the multi-dimensional data to obtain spectrum data.

[0176] Step S222, input the spectrum data into the pre-built nitrification machine fault warning model to obtain a fault warning result, wherein the fault diagnosis result includes fault characteristics and fault causes.

[0177] Furthermore, the nitrification machine fault warning model includes a beta-variational autoencoder module, an error distribution module and a comparison module connected in sequence. The beta-variational autoencoder module is used to perform spectrum reconstruction on the spectrum data using the trained beta-variational autoencoder model to obtain spectrum reconstruction data. The error distribution module is used to calculate the error distribution of the spectrum reconstruction data to obtain first error distribution data. The comparison module is used to compare the first error distribution data with a predetermined first warning threshold, and use the comparison result as the fault warning result.

[0178] Specifically, this embodiment uses the spectrum obtained by Fourier transform combined with the beta-Variational Autoencoder (beta-VAE) model in deep learning to perform intelligent detection of the "difference" of the fault characteristic signal of the nitrifier to achieve intelligent early warning and identify the real-time data characteristic values ​​of the nitrifier and the abnormal conditions of the original waveform.

[0179] Preferably, the trained beta-variational autoencoder model is determined by the following steps (a) to (c). Step (a), obtaining fault warning training sample data; wherein the fault warning training sample data is obtained by simulating the normal working state of the nitrifier by a pre-built nitrifier fault simulation test bench. Step (b), performing a second preprocessing on the fault warning training sample data to obtain spectrum training data; wherein the second preprocessing includes normalization and Fourier transform. Step (c), inputting the spectrum training data into a pre-built beta-variational autoencoder model for training to obtain a trained beta-variational autoencoder model and spectrum reconstruction training data.

[0180] Step (b) normalizes and Fourier transforms the original data of the nitrifier, and uses the spectrum of the nitrifier under normal working conditions as input. A beta-VAE model is pre-built and trained based on the above input data to achieve reconstruction of the health status monitoring spectrum of the nitrifier, calculate the error distribution of the data, and determine the warning threshold through the error range. Finally, the field data and the fault simulation test bench data are predicted, and then the predicted result error and the warning threshold are compared to confirm whether the nitrifier is in normal working condition. The normal data identified by the beta-VAE model is added to the input data set to expand the sample size, so as to more accurately calculate the error distribution and determine the warning threshold of the nitrifier, thereby minimizing the number of false alarms while ensuring that no faults of the nitrifier are missed.

[0181] Taking the vibration data of a nitrifier as an example, a neural network based on a 1D convolutional layer is used for preprocessing. The specific parameters of the network are as follows: Activation function: SeLU activation function; Encoder convolution hidden layer: 6 layers, the number of channels in each layer: 4, 8, 16, 32, 16, 8, and the dimension is reduced to 1 / 2 of the original length after each layer, Encoder fully connected layer: 2 layers, the number of neurons: 64, 20 (twice the number of hidden variables); Decoder hidden layer: 8 groups, each group contains a deconvolution layer and a feature fusion layer, the number of deconvolution channels in each layer: 64, 32, 32, 16, 16, 8, 4, 2; Total number of trainable parameters: 228k; Optimizer selection: Adam; Training step size: 0.0005, number of iterations: 10k, initial C value: 0.5^7.

[0182] The reconstruction effect is as follows Figure 8 As shown in the figure, curve S1 represents the original spectrum, and curve S2 represents the reconstructed spectrum. The model also has a very low reconstruction error outside the training set, and its error rate is basically consistent with the previous period. Even if the spectrum changes slightly, the reconstruction model can still reconstruct well. However, around the 2300th data set (minutes), the reconstruction error increases significantly, which is caused by a fault. It can be seen that a fault will cause an increase in the reconstruction error. Figure 8 In the figure, the reconstruction error trend over time, where the first 1000 groups are training sets. As can be seen from the green box, the reconstruction error did not exceed the healthy level until around 2300 groups.

[0183] The fault warning based on the beta-variational autoencoder model of the present embodiment adopts the beta-VAE model in the spectrum combined with deep learning to perform intelligent detection of the "difference" of the fault characteristic signal of the nitrifier to achieve intelligent warning, and identify the real-time data characteristic values ​​of the nitrifier and the abnormal conditions of the original waveform; the fault warning based on the beta-variational autoencoder model of the present embodiment can not only effectively extract the deep features in the fault signal of the nitrifier, but also enhance the model's adaptability to unseen faults when the sample size is limited, thereby improving the intelligent level of nitrifier maintenance and enhancing the efficiency and accuracy of fault prevention, which is of great significance for ensuring the continuity and safety of industrial production.

[0184] Here, the health status of the nitrifier is evaluated according to the multi-dimensional data to obtain the health status evaluation result, which specifically includes the following steps: Step S230, inputting the multi-dimensional data into the health status evaluation model of the nitrifier constructed in advance based on the nitrifier index system, scoring the health status of the nitrifier, and obtaining the health status evaluation result. Among them, the index system is constructed based on the status and structural composition of the nitrifier; the status includes the operation status, fault and maintenance status and environmental status of the nitrifier; the structural composition includes the shaft, bearing, motor and agitator.

[0185] The key components are extracted from the structure of the nitrifier to construct the first-level indicators, including shafts, bearings, motors and agitators; the second-level indicators are constructed from the three dimensions of operation, failure and maintenance, and environmental conditions, and finally a health index of 0-100 is calculated. The mean time between failures is used to reflect the frequency of failures in the nitrifier in the failure and maintenance conditions; the environmental conditions refer to the ambient temperature and humidity of the nitrifier; the motor operation indicators include motor current, motor speed, and operating time; the shaft operation indicators include shaft temperature, shaft wear, and operating time; the bearing operation indicators include bearing temperature and bearing vibration; the agitator operation indicators include agitator speed, agitator power, and operating time.

[0186] The steps of health status assessment are as follows: Fig. 9As shown, the following steps are included: Step 710, establishing a health assessment system for key equipment of the nitrifier; Step 720, calculating the weight of each indicator by combining the analytic hierarchy process and the entropy weight method; Step 730, establishing a scoring standard for each indicator; Step 740, equipment health assessment score = weight of each indicator × sum of the scores of each indicator.

[0187] (1) Operation status (calculated daily from the time the nitrification equipment is put into use). The operation status includes the operation time score and the degree of degradation; the failure and maintenance status includes the mean time between failures score; the environmental conditions include the environmental temperature score and the environmental humidity score. The specific indicators contained in the operation status are able to characterize the process of the performance or health status of the key equipment of the nitrification machine gradually deteriorating over time, and are calculated from the time the equipment is put into use.

[0188] ① The operating time score is determined by the following formula (1). The service life parameter is calculated daily based on the time when the key equipment of the nitrifier is put into use and the time when the key equipment of the nitrifier is in operation.

[0189]

[0190] In formula (1), YE represents the operating time score; x1 represents the number of years that the key equipment of the nitrification machine has been put into use; k1 represents the design life of the key equipment of the nitrification machine.

[0191] ② Degradation degree (abnormal situation): determine the monitoring indicators of the equipment through risk analysis of key equipment of the nitrifier and sensor layout, and determine the abnormal operation of the equipment based on the monitoring values ​​of these indicators.

[0192] The degradation degree is determined by the following formula (2) and formula (3), wherein, for the case where the state detection parameters and performance parameters can be used to reflect the equipment state, for the characteristic parameter with the shutdown parameter as an upper limit, its degradation degree can be calculated by the following formula (2); for the characteristic parameter with the shutdown parameter as a lower limit, its degradation degree can be calculated by the following formula (3).

[0193]

[0194] In formula (2) and formula (3), d j Indicates the degree of degradation; x i represents the actual measured value of the characteristic parameters of the key equipment of the nitrification machine; f Indicates the normal value or factory value of the characteristic parameters of the key equipment of the nitrifier; x max Indicates the upper limit characteristic parameter threshold when the key equipment of the nitrification machine needs to be shut down; x min Indicates the lower limit characteristic parameter threshold when the key equipment of the nitrification machine is shut down; k2 represents x i The relationship between k2 and the health status of the key equipment of the nitrifier is determined by experience. When k2 is 1, the xi It is linearly related to the nitrifier or the key equipment of the nitrifier, that is, the greater the degree of degradation, the worse the health status of the key equipment of the nitrifier.

[0195] When the characteristic parameter threshold x max and x min When it is a range threshold, the degradation degree is determined by the following formula (4).

[0196]

[0197] In formula (4), x1 and x2 respectively represent the lower limit and upper limit of the characteristic parameters of the key equipment of the nitrification machine when the condition is normal, and x0 and x3 respectively represent the lower limit and upper limit of the characteristic parameters when the key equipment of the nitrification machine must be shut down.

[0198] (2) Failure and maintenance. The mean time between failures (MTBF) reflects the average time during which failures occur during the entire operation of the equipment, including failure time and maintenance time, and reflects the frequency of equipment failures. The larger the MTBF, the higher the reliability of the equipment and the stronger its ability to work correctly. In actual calculations, the score is calculated based on the failure level. The failure level is determined by the failure mode and effects analysis (FMEA) method, and the proportion of each level of failure can be determined by the weight of the hierarchical analysis method.

[0199] When the nitrification equipment is in a faulty state, the nitrification equipment is alarming and has not recovered, and the score of this equipment is 0. When there is no alarm, the score refers to MTBF. The specific algorithm is MTBF (time / time) = total operating time / total number of faults. The inverse of MTBF, that is, the failure rate, is taken to calculate the score.

[0200] The mean time between failures score is determined by the following formula (5).

[0201]

[0202] In formula (5), M1 represents the mean time between failures score; MTBF represents the mean time between failures, which is determined by the ratio of the total operating time of the nitrifier to the total number of failures.

[0203] (3) Environmental conditions. Ambient temperature and humidity are constantly changing and are calculated once a day.

[0204] ① Ambient temperature. The operating environment is composed of temperature and humidity. The normal equipment environment temperature is defined as X℃-Y℃. The current equipment operating environment temperature score is defined as 0 points for ≤X4℃ and ≥Y4℃, and ((X4+Y4) / 2)℃ for 100 points. The ambient temperature score is determined by the following formula (6).

[0205]

[0206] In formula (6), T represents the ambient temperature score, t2 represents the current ambient temperature; the normal range threshold of ambient temperature is defined as X4℃~Y4℃; , T=100.

[0207] ② Ambient humidity. The absolute humidity of the equipment environment is defined as M2g / m 3 ~N2g / m 3 The absolute humidity score of the current equipment operating environment is defined as ≤M2g / m 3 and ≥N2g / m 3 0 points, ((M2+N2) / 2)g / m 3 The score of ambient temperature is 100. The score of ambient temperature is determined by formula (7).

[0208]

[0209] In formula (7), H represents the environmental humidity score, t3 represents the current environmental humidity; the normal range threshold of environmental absolute humidity is defined as M2g / m 3 ~N2g / m 3 Among them, when , H=100.

[0210] The health status evaluation system of the nitrification machine in this embodiment is constructed by first extracting key components from the structural composition of the nitrification machine as primary indicators, and then constructing secondary indicators from three dimensions: operation status, failure and maintenance status, and environmental status.

[0211] It should be noted that, in actual applications, the fault diagnosis step S210, the fault warning step S220 and the health status assessment step S230 are not performed sequentially. The fault warning step is performed in real time, while the fault diagnosis step is generally performed after the fault occurs, and the health status assessment step is performed periodically.

[0212] Step S300, displaying fault diagnosis results, fault warning results and health status assessment results.

[0213] Reference Figure 4 This embodiment proposes a full-process technical framework with nitrification machine as the research object, including state monitoring parameter selection, multi-dimensional data collection, weak fault feature extraction, mechanism and data dual-driven nitrification machine fault diagnosis, fault warning based on beta-variational autoencoder model and health status assessment system construction. This embodiment fills the gap of systematic solutions in the field of state monitoring and health management of nitrification machine. As part of intelligent manufacturing, the promotion and application of this framework will accelerate the intelligent transformation of fire-related enterprises and promote technological progress and industrial upgrading.

[0214] The method of this embodiment is applied to a fault simulation test bench and an actual production environment of a nitrification machine to verify the actual application effect of the above technical framework.

[0215] This embodiment proposes a mechanism- and data-driven nitrification machine fault diagnosis method, which not only establishes a fault diagnosis rule for the nitrification machine based on the fault mechanism of the equipment, but also constructs a data-driven fault diagnosis model by combining a one-dimensional convolutional neural network (1DCNN), an improved gated recurrent unit neural network (GGRU) and an attention mechanism. Finally, the fault category obtained by diagnosis is matched to a fault mode library constructed based on the mechanism to obtain detailed information including fault characteristics and fault causes, thereby solving the problem of lack of interpretability of a fault diagnosis model based solely on deep learning technology. The fault diagnosis accuracy reaches more than 85%, which significantly improves the initiative and efficiency of equipment maintenance.

[0216] This embodiment proposes a set of health status assessment methods, which provide a scientific basis for the formulation of nitrification machine maintenance plans by continuously monitoring and evaluating the health status of the nitrification machine, and promote the transformation of nitrification machine maintenance from "after-the-fact maintenance" to "predictive maintenance".

[0217] In practical applications, the above innovations effectively reduce production interruptions caused by downtime due to faults; through accurate prediction and early intervention, the number of unplanned downtimes and maintenance costs of nitrification machines are reduced, saving annual maintenance costs for fire-related enterprises; the utilization rate and stability of nitrification machines are improved, the production delay time caused by faults is shortened, and the overall production efficiency is improved. In terms of social value, the application of this embodiment can timely detect and warn of nitrification machine failures, provide support for the safety of production personnel and equipment, and provide a strong guarantee for the smooth operation of fire-related enterprises.

[0218] In summary, the method of this embodiment covers the technical architecture of equipment health status monitoring parameter selection, multi-dimensional data collection, weak fault feature extraction, fault diagnosis, fault warning, health status assessment, and technical verification. In view of the limitations of existing equipment fault diagnosis methods, this embodiment proposes a nitrification machine fault diagnosis model that combines deep learning with attention mechanism, which effectively integrates data and fault mechanism, improves the accuracy of fault diagnosis and the interpretability of diagnosis results; proposes a health status assessment model that integrates fault mechanism, which provides a reference for the health assessment of fire-related equipment. This embodiment can realize the unmanned operation of dangerous places and the real-time monitoring of the health status of nitrification machines, improve the inherent safety of nitrification machines, and is conducive to the promotion of the work of "human-machine isolation, machine replacement, and dark factory".

[0219] See also Fig.10This embodiment provides a multi-dimensional state monitoring and health management device for a nitrifier, including a data acquisition module 100, an intelligent diagnosis and early warning module 200, and a display module 300. The data acquisition module 100 is used to acquire multi-dimensional data of the nitrifier; the multi-dimensional data at least includes vibration data, current data, and temperature data of the nitrifier. The intelligent diagnosis, early warning, and health status assessment module 200 is used to perform fault diagnosis on the nitrifier according to the multi-dimensional data to obtain a fault diagnosis result; and / or perform fault early warning on the nitrifier according to the multi-dimensional data to obtain a fault early warning result; and / or assess the health status of the nitrifier according to the multi-dimensional data to obtain a health status assessment result of the nitrifier. The display module 300 is used to display the fault diagnosis result, and / or the fault early warning result, and / or the health status assessment result.

[0220] Among them, see Fig.10 The monitoring and diagnosis module 200 includes a fault diagnosis module 210, a fault warning module 220 and a health status assessment module 230. The fault diagnosis module 210 includes a first processing module 211, a diagnosis module 212 and a first result module 213. The first processing module 211 is used to perform a first preprocessing and feature extraction on the multi-dimensional data to obtain a multi-dimensional data feature set of the nitrification machine. The diagnosis module 212 is used to input the multi-dimensional data feature set of the nitrification machine into a trained nitrification machine fault diagnosis model to obtain a fault category label. The first result module 213 is used to input the fault category label into a pre-built fault mode library to obtain a fault diagnosis result of the nitrification machine. Optionally, the fault diagnosis result includes fault characteristics and fault causes.

[0221] See also Fig.10 The fault warning module 220 includes a second processing module 221 and a second result module 222. The second processing module 221 is used to perform a second preprocessing on the multi-dimensional data to obtain spectrum data. The second result module 222 is used to input the spectrum data into a pre-built nitrification machine fault warning model to obtain a fault warning result.

[0222] In an optional embodiment, the health status assessment module 230 is used to input multi-dimensional data into a nitrifier health status assessment model that is pre-constructed based on a nitrifier index system, to score the health status of the nitrifier, and to obtain a health status assessment result; wherein the index system is constructed based on the status and structural composition of the nitrifier; the status includes the operating status, fault and maintenance status, and environmental status of the nitrifier; and the structural composition includes a shaft, bearings, a motor, and an agitator.

[0223] The trained nitrification machine fault diagnosis model is obtained through the following sequentially connected modules: a first sample acquisition module, a first sample preprocessing module, a first feature set module and a first training module. The first sample acquisition module is used to obtain fault diagnosis training sample data. The first sample preprocessing module is used to perform a first preprocessing on the fault diagnosis training sample data to obtain the fault diagnosis training sample data after the first preprocessing. The first feature set module is used to extract features from the fault diagnosis training sample data after the first preprocessing to obtain a nitrification machine training feature set. The first training module is used to input the nitrification machine training feature set into a pre-constructed nitrification machine fault diagnosis model for training to obtain a trained nitrification machine fault diagnosis model; wherein the nitrification machine fault diagnosis model includes an improved gated recurrent unit network module, a one-dimensional convolutional neural network module, an attention mechanism module and a Softmax classifier module, the improved gated recurrent unit network module and the one-dimensional convolutional neural network module are respectively connected to the Softmax classifier module, and the attention mechanism module is connected to the improved gated recurrent unit network module.

[0224] The improved gated recurrent unit network module is obtained by sequentially connecting the following modules: a construction module, a GGRU network primary module and a GGRU network module. The construction module is used to construct the gated recurrent unit network module. The GGRU network primary module is used to replace the output layer of the gated recurrent unit network module with a one-dimensional global pooling layer to obtain an improved gated recurrent unit network primary module. The GGRU network module is used to adjust the output layer of the improved gated recurrent unit network primary module by using the attention mechanism module to obtain an improved gated recurrent unit network module.

[0225] In an optional embodiment, the fault diagnosis training sample data includes first fault diagnosis training sample data and second fault diagnosis training sample data; wherein, the first fault diagnosis training sample data is obtained by performing fault simulation on a pre-built nitrification machine fault simulation test bench; the second fault diagnosis training sample data is constructed using a pre-built interference model and normal samples; and the normal samples are obtained by simulating the normal working state of the nitrification machine on the nitrification machine fault simulation test bench.

[0226] In an optional embodiment, the feature set module includes a weak signal feature extraction module, a multidimensional feature extraction module and a feature set construction module. The weak signal feature extraction module is used to extract the weak signal in the fault diagnosis training sample data after the first preprocessing using the ARLW algorithm, and to enhance the weak signal using the SPM algorithm to obtain the enhanced weak signal; the enhanced weak signal is impact quantized using the Hilbert envelope demodulation algorithm and the peak energy value to obtain weak signal feature data. The multidimensional feature extraction module is used to extract multidimensional feature data in the fault diagnosis training sample data after the first preprocessing; wherein the multidimensional feature data includes time domain features, frequency domain features and time-frequency domain features. The feature set construction module is used to construct a nitrification machine training feature set using weak signal feature data and multidimensional feature data.

[0227] In an optional embodiment, the nitrification machine fault warning model includes a beta-variational autoencoder module, an error distribution module and a comparison module connected in sequence. The beta-variational autoencoder module is used to perform spectrum reconstruction on the spectrum data using the trained beta-variational autoencoder model to obtain spectrum reconstruction data. The error distribution module is used to calculate the error distribution of the spectrum reconstruction data to obtain first error distribution data. The comparison module is used to compare the first error distribution data with a predetermined first warning threshold, and use the comparison result as the fault warning result.

[0228] In an optional embodiment, the trained beta-variational autoencoder model is determined by the following sequentially connected modules: a second sample acquisition module, a second sample preprocessing module, and a second training module. The second sample acquisition module is used to obtain fault warning training sample data; wherein, the fault warning training sample data is obtained by simulating the normal working state of the nitrifier through a pre-built nitrifier fault simulation test bench. The second sample preprocessing module is used to perform a second preprocessing on the fault warning training sample data to obtain spectrum training data; wherein, the second preprocessing includes Fourier transform. The second training module is used to input the spectrum training data into the pre-built beta-variational autoencoder model for training, and obtain a trained beta-variational autoencoder model and spectrum reconstruction training data.

[0229] In an optional embodiment, the operating conditions include an operating time score and a degree of degradation; the fault and maintenance conditions include a mean time between failures score; and the environmental conditions include an environmental temperature score and an environmental humidity score.

[0230] The running time score is determined by the following formula (1).

[0231]

[0232] In formula (1), YE represents the operating time score; x1 represents the number of years that the key equipment of the nitrifier has been put into use; k1 represents the design life of the key equipment of the nitrifier.

[0233] The degradation degree is determined by the following equations (2) and (3):

[0234]

[0235] In formula (2) and formula (3), d j Indicates the degree of degradation; x i represents the actual measured value of the characteristic parameters of the key equipment of the nitrification machine; f Indicates the normal value or factory value of the characteristic parameters of the key equipment of the nitrifier; x max Indicates the upper limit characteristic parameter threshold when the key equipment of the nitrification machine needs to be shut down; x min Indicates the lower limit characteristic parameter threshold when the key equipment of the nitrification machine is shut down; k2 represents x i The relationship between the health status of the key equipment of the nitrifier and the i It is linearly related to the nitrification machine or key equipment of the nitrification machine;

[0236] When the characteristic parameter threshold x max and x min When is the range threshold, the degradation degree is determined by the following formula (4):

[0237]

[0238] In formula (4), x1 and x2 represent the lower limit and upper limit of the characteristic parameters of the key equipment of the nitrification machine when they are normal, and x0 and x3 represent the lower limit and upper limit of the characteristic parameters when the key equipment of the nitrification machine must be shut down;

[0239] The mean time between failures score is determined by the following formula (5):

[0240]

[0241] In formula (5), M1 represents the mean time between failures score; MTBF represents the mean time between failures, which is determined by the ratio of the total operating time of the nitrifier to the total number of failures;

[0242] The ambient temperature score is determined by the following formula (6):

[0243]

[0244] In formula (6), T represents the ambient temperature score, t2 represents the current ambient temperature; the normal range threshold of ambient temperature is defined as X4℃~Y4℃; When T=100;

[0245] The ambient humidity score is determined by the following formula (7):

[0246]

[0247] In formula (7), H represents the environmental humidity score, t3 represents the current environmental humidity; the normal range threshold of environmental absolute humidity is defined as M2g / m 3 ~N2g / m 3 Among them, when , H=100.

[0248] The device provided in the embodiment of the present application is adopted. Since the device adopts the same inventive concept as the above-mentioned method provided in the embodiment of the present application, on the premise that the method can solve the technical problem, the device can also solve the technical problem, which will not be repeated here.

[0249] It is known from common technical knowledge that the present invention can be implemented by other embodiments that do not deviate from its spirit or essential features. Therefore, the above disclosed embodiments are only illustrative in all respects and are not exclusive. All changes within the scope of the present invention or within the scope equivalent to the present invention are included in the present invention.

Claims

1. A multi-dimensional state monitoring and health management method for a nitrification machine, characterized in that: include: Select the nitrification machine status monitoring parameters to obtain multi-dimensional data during the operation of the nitrification machine; The multi-dimensional data at least includes vibration data, current data and temperature data of the nitrification machine; Performing fault diagnosis on the nitrification machine according to the multi-dimensional data to obtain a fault diagnosis result; and / or performing fault warning on the nitrification machine according to the multi-dimensional data to obtain a fault warning result; and / or, performing a health status assessment on the nitrification machine according to the multidimensional data to obtain a health status assessment result; Display the fault diagnosis results, and / or fault warning results, and / or health status assessment results; Wherein, the fault diagnosis of the nitrification machine is performed according to the multi-dimensional data to obtain the fault diagnosis result, including: Performing a first preprocessing and feature extraction on the multi-dimensional data to obtain a multi-dimensional data feature set of the nitrification machine; Inputting the multi-dimensional data feature set of the nitrification machine into the trained nitrification machine fault diagnosis model to obtain a fault category label; Inputting the fault category label into a pre-built fault mode library to obtain a fault diagnosis result of the nitrification machine; The method of performing a fault warning on the nitrification machine according to the multi-dimensional data to obtain a fault warning result includes: Performing a second preprocessing on the multi-dimensional data to obtain spectrum data; Inputting the spectrum data into a pre-built nitrification machine fault warning model to obtain a fault warning result; The health status evaluation of the nitrification machine is performed according to the multi-dimensional data to obtain a health status evaluation result, including: The multi-dimensional data is input into a nitrification machine health status assessment model pre-constructed based on a nitrification machine index system, and a health status score is performed on the nitrification machine to obtain a health status assessment result.

2. The multi-dimensional state monitoring and health management method of the nitrification machine according to claim 1 is characterized in that: The trained nitrifier fault diagnosis model is obtained by the following method: Obtain fault diagnosis training sample data; Performing a first preprocessing on the fault diagnosis training sample data to obtain the first preprocessed fault diagnosis training sample data; Performing feature extraction on the fault diagnosis training sample data after the first preprocessing to obtain a nitrification machine training feature set; The nitrification machine training feature set is input into a pre-built nitrification machine fault diagnosis model for training to obtain a trained nitrification machine fault diagnosis model; wherein the nitrification machine fault diagnosis model includes an improved gated recurrent unit network module, a one-dimensional convolutional neural network module, an attention mechanism module and a Softmax classifier module, the improved gated recurrent unit network module and the one-dimensional convolutional neural network module are respectively connected to the Softmax classifier module, and the attention mechanism module is connected to the improved gated recurrent unit network module; The improved gated recurrent unit network module is obtained by the following method: Construct a gated recurrent unit network module; Using a one-dimensional global pooling layer to replace the output layer of the gated recurrent unit network module to obtain an improved gated recurrent unit network primary module; The attention mechanism module is used to adjust the output layer of the improved gated recurrent unit network primary module to obtain an improved gated recurrent unit network module.

3. The multi-dimensional state monitoring and health management method of the nitrification machine according to claim 2 is characterized in that: The fault diagnosis training sample data includes first fault diagnosis training sample data and second fault diagnosis training sample data; wherein, the first fault diagnosis training sample data is obtained by performing fault simulation on a pre-built nitrification machine fault simulation test bench; the second fault diagnosis training sample data is constructed using a pre-built interference model and normal samples; the normal samples are obtained by simulating the normal working state of the nitrification machine on the nitrification machine fault simulation test bench.

4. The multi-dimensional state monitoring and health management method of the nitrification machine according to claim 2 is characterized in that: The feature extraction of the fault diagnosis training sample data after the first preprocessing is performed to obtain a nitrification machine training feature set, including: A weak signal is extracted from the fault diagnosis training sample data after the first preprocessing by using the ARLW algorithm, and the weak signal is enhanced by using the SPM algorithm to obtain an enhanced weak signal; the enhanced weak signal is impact quantized by using the Hilbert envelope demodulation algorithm and the peak energy value to obtain weak signal feature data; and multidimensional feature data is extracted from the fault diagnosis training sample data after the first preprocessing; wherein the multidimensional feature data includes time domain features, frequency domain features and time-frequency domain features; The weak signal feature data and the multi-dimensional feature data are used to construct the nitrification machine training feature set.

5. The multi-dimensional state monitoring and health management method of the nitrification machine according to claim 1 is characterized in that: The nitrification machine fault warning model includes a beta-variational autoencoder module, an error distribution module and a comparison module connected in sequence; The beta-variational autoencoder module is used to perform spectrum reconstruction on the spectrum data using a trained beta-variational autoencoder model to obtain spectrum reconstructed data; The error distribution module is used to calculate the error distribution of the spectrum reconstruction data to obtain first error distribution data; The comparison module is used to compare the first error distribution data with a predetermined first warning threshold, and use the comparison result as a fault warning result.

6. The multi-dimensional state monitoring and health management method of the nitrification machine according to claim 5 is characterized in that: The trained beta-variational autoencoder model is determined by the following method: Acquire fault warning training sample data; wherein the fault warning training sample data is obtained by simulating the normal working state of the nitrifier through a pre-built nitrifier fault simulation test bench; Performing a second preprocessing on the fault warning training sample data to obtain spectrum training data; wherein the second preprocessing includes Fourier transform; The spectrum training data is input into a pre-built beta-variational autoencoder model for training to obtain a trained beta-variational autoencoder model and spectrum reconstruction training data.

7. The multi-dimensional state monitoring and health management method of the nitrification machine according to claim 1 is characterized in that: The method further comprises: The index system is constructed based on the status and structural composition of the nitrifier; the status includes the operation, failure and maintenance, and environmental conditions of the nitrifier; and the structural composition includes a shaft, bearings, a motor, and an agitator.

8. The multi-dimensional state monitoring and health management method of the nitrification machine according to claim 7 is characterized in that: The operation condition includes the operation time score and the degree of degradation; the failure and maintenance condition includes the mean time between failures score; the environmental condition includes the environmental temperature score and the environmental humidity score; The running time score is determined by the following formula (1): In formula (1), YE represents the operating time score; x1 represents the number of years that the key equipment of the nitrification machine has been put into use; k1 represents the design life of the key equipment of the nitrification machine; The degradation degree is determined by the following equations (2) and (3): In formula (2) and formula (3), d j Indicates the degree of degradation; x i represents the actual measured value of the characteristic parameters of the key equipment of the nitrification machine; f Indicates the normal value or factory value of the characteristic parameters of the key equipment of the nitrification machine; x max Indicates the upper limit characteristic parameter threshold when the key equipment of the nitrification machine needs to be shut down; x min represents the lower limit characteristic parameter threshold when the key equipment of the nitrification machine is shut down; k2 represents x i The relationship between the health status of the key equipment of the nitrification machine and the i It is linearly related to the nitrification machine or key equipment of the nitrification machine; When the characteristic parameter threshold x max and x min When is a range threshold, the degradation degree is determined by the following formula (4): In formula (4), x1 and x2 respectively represent the lower limit and upper limit of the characteristic parameters of the key equipment of the nitrification machine when the parameters are normal, and x0 and x3 respectively represent the lower limit and upper limit of the characteristic parameters when the key equipment of the nitrification machine must be shut down; The mean time between failures score is determined by the following formula (5): In formula (5), M1 represents the mean time between failures score; MTBF represents the mean time between failures, and the mean time between failures is determined by the ratio of the total operating time of the nitrifier to the total number of failures; The ambient temperature score is determined by the following formula (6): In formula (6), T represents the ambient temperature score, t2 represents the current ambient temperature; the normal range threshold of ambient temperature is defined as X4℃~Y4℃; When T=100; The environmental humidity score is determined by the following formula (7): In formula (7), H represents the environmental humidity score, t3 represents the current environmental humidity; the normal range threshold of environmental absolute humidity is defined as M2g / m 3 ~N2g / m 3 Among them, when , H=100.

9. The multi-dimensional state monitoring and health management method of a nitrification machine according to claim 1, characterized in that: The fault diagnosis result includes fault characteristics and fault causes.

10. A multi-dimensional state monitoring and health management device for a nitrification machine, characterized in that: include: A data acquisition module is used to obtain multi-dimensional data of the nitrification machine; The multi-dimensional data at least includes vibration data, current data and temperature data of the nitrification machine; An intelligent diagnosis and early warning module is used to perform fault diagnosis on the nitrification machine according to the multi-dimensional data to obtain a fault diagnosis result; and / or to perform fault early warning on the nitrification machine according to the multi-dimensional data to obtain a fault early warning result; and / or, performing a health status assessment on the nitrification machine according to the multidimensional data to obtain a health status assessment result; A display module, used to display the fault diagnosis result, and / or the fault warning result, and / or the health status assessment result; Wherein, the intelligent diagnosis and early warning module includes a fault diagnosis module, and / or a fault early warning module, and / or a health status assessment module; The fault diagnosis module comprises: A first processing module, used for performing a first preprocessing and feature extraction on the multi-dimensional data to obtain a multi-dimensional data feature set of the nitrification machine; A diagnosis module, used for inputting the multi-dimensional data feature set of the nitrification machine into a trained nitrification machine fault diagnosis model to obtain a fault category label; A first result module is used to input the fault category label into a pre-built fault mode library to obtain a fault diagnosis result of the nitrification machine; The fault warning module comprises: A second processing module, used for performing a second preprocessing on the multi-dimensional data to obtain spectrum data; The second result module is used to input the spectrum data into a pre-built nitrification machine fault warning model to obtain a fault warning result; The health status assessment module is used to input the multi-dimensional data into a nitrification machine health status assessment model pre-constructed based on a nitrification machine index system, perform a health status score on the nitrification machine, and obtain a health status assessment result.

Images