CN109779894B - Reciprocating compressor fault diagnosis system and method based on neural network algorithm - Google Patents
Reciprocating compressor fault diagnosis system and method based on neural network algorithm Download PDFInfo
- Publication number
- CN109779894B CN109779894B CN201811643547.8A CN201811643547A CN109779894B CN 109779894 B CN109779894 B CN 109779894B CN 201811643547 A CN201811643547 A CN 201811643547A CN 109779894 B CN109779894 B CN 109779894B
- Authority
- CN
- China
- Prior art keywords
- compressor
- fault
- simulation
- neural network
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Landscapes
- Control Of Positive-Displacement Pumps (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a fault diagnosis system and method of a reciprocating compressor based on a neural network algorithm. The method comprises the steps of obtaining a simulation model of the compressor under each working condition by utilizing a neural network algorithm and combining actually measured state parameters of the compressor system; the method comprises the steps that a standard fault characteristic is obtained by simulating fault calculation of a compressor through changing system parameters, and after the compressor runs abnormally, the fault position can be accurately positioned and a removing method can be provided through comparison with the standard fault characteristic; the invention has high diagnosis success rate, does not need high-precision equipment for real-time monitoring and analysis, and reduces the diagnosis cost.
Description
Technical Field
The invention relates to the technical field of equipment fault diagnosis, in particular to a fault diagnosis system and method of a reciprocating compressor based on a neural network algorithm.
Background
The reciprocating compressor is used as core equipment for oil and gas field transportation, and the stability, reliability and safety of the reciprocating compressor are of great significance. Because the compressor has a complex structure, more parts and complex assembly, sometimes the reciprocating compressor produced by using the same drawing has different operating characteristics in field operation, so that an accurate integral mathematical simulation model is difficult to construct for the compressor, and how to reduce direct or indirect loss caused by the compressor fault through fault diagnosis when a fault symptom appears is always an important subject in the compressor industry.
At present, vibration detection, thermal parameter detection, stress strain detection and the like are mainstream detection methods, however, the above detection only detects and judges specific faults of the compressor, and cannot compare and judge the overall state and the system state of the compressor. At present, some fault diagnosis systems of engine equipment monitor dynamic and vibration characteristic parameters of an engine in real time on line for fault diagnosis, but because the thermodynamic model is not corrected and a method for monitoring a fault characteristic spectrum in a time-frequency domain signal is adopted, false alarm is easy to occur, and finally, most of technicians can only process the reasons causing the fault by means of experience, and time and labor are consumed, and errors are easy to occur.
Disclosure of Invention
The technical problem to be solved by the present invention is to provide a fault diagnosis system and method for a reciprocating compressor based on a neural network algorithm, which can accurately locate a fault and provide a method for removing the fault, aiming at the defects existing in the prior art.
The invention is realized by the following technical scheme:
a reciprocating compressor fault diagnosis system based on a neural network algorithm comprises a field communication module, a data acquisition module, a simulation module based on a neural network, an expert system module based on standard fault characteristics, a fault diagnosis host and an upper computer;
the data acquisition module is used for acquiring characteristic parameters of each part of the compressor and sending the characteristic parameters to the simulation module based on the neural network;
the field communication module is used for acquiring real-time operation parameters of the compressor from the PLC monitoring system and sending the real-time operation parameters to the simulation module based on the neural network;
the simulation module based on the neural network establishes a mathematical simulation model of the compressor according to the design data of the compressor, the characteristic parameters of each part acquired by the data acquisition module during the normal working period of the compressor and the real-time operation parameters acquired by the field communication module and stores the mathematical simulation model to the expert system module; various faults of different parts of the compressor are simulated by changing parameters of relevant fault parts in the mathematical simulation model to obtain fault data characteristics and the fault data characteristics are stored in an expert system module;
the expert system module comprises a comprehensive database and an expert knowledge base, wherein the comprehensive database comprises a mathematical simulation model and fault data characteristics of the compressor; the expert knowledge base comprises a fault judgment criterion and a processing method aiming at various faults;
the fault diagnosis host is used for reading fault diagnosis processes, results and elimination measures by a user;
and the upper computer is used for a developer to read, add and modify the simulation module and the expert system module based on the neural network.
Preferably, the data acquisition module comprises a temperature sensor, a pressure sensor, a strain gauge sensor, a vibration sensor and an eddy current sensor.
Preferably, the expert system module further comprises an inference engine, and the inference engine is used for searching for an applicable fault judgment criterion by using the control strategy.
Preferably, the integrated database and the expert knowledge base can be read, added and modified.
Preferably, the field communication module supports Modbus protocol, TCP/IP protocol and CAN bus communication protocol.
A fault diagnosis method for a reciprocating compressor based on a neural network algorithm comprises the following steps:
s1, acquiring characteristic parameters of each part of the compressor through the data acquisition module, and sending the characteristic parameters to the simulation module based on the neural network;
s2, acquiring real-time operation parameters of the compressor PLC monitoring system through the field communication module, and sending the real-time operation parameters to the simulation module based on the neural network;
s3, carrying out simulation by a simulation module based on the neural network: in the compressor assembling and trial run stage, a simulation module of the neural network adopts compressor design data, characteristic parameters of each part of the compressor and real-time operation parameters of the compressor, which are acquired by a data acquisition module, and establishes an ideal thermodynamic equation, an ideal kinetic equation and an ideal vibration mechanical equation of the compressor after filtering and removing noise to obtain an ideal mathematical model of the compressor; in the loading operation stage of the compressor, real-time operation parameters acquired by a field communication module are input into an ideal mathematical model, a simulation thermodynamic model, a simulation kinetic model and a simulation vibration mechanical model of the compressor are established by combining a BP neural network method and structural dynamics modification, the simulation mathematical model of the compressor is obtained, and the established simulation mathematical model is stored in an expert system module; iteratively changing different parameters of relevant fault parts in the simulation mathematical model through a neural network algorithm, simulating various faults of different parts of the compressor to obtain fault data characteristics, and storing the fault data characteristics to an expert system module;
s4, fault diagnosis: when a fault symptom occurs, the real-time operation parameters of the compressor acquired by the field communication module are used as fault data characteristics to be matched, and the fault data characteristics to be matched and the fault data characteristics stored in the expert system module are judged to be matched according to a fault judgment criterion in the expert knowledge base; if the matching is successful, fault diagnosis is realized, and corresponding expert opinions are given; if the matching is unsuccessful, relevant fault part parameters of the compressor simulation mathematical model are modified through the neural network algorithm in an iterative mode until the simulated fault data characteristics are successfully matched with the fault data characteristics to be matched, then a fault elimination method in an expert knowledge base is called out, fault diagnosis is completed, and the simulated fault data characteristics are stored in an expert system module.
Preferably, step S1 specifically includes: when the unit is machined, the air cylinder, the middle body, the machine body, the cylinder head support, the middle body support, the buffer tank and the washing tank are tested through the data acquisition module, the natural frequency, the modal vibration mode and the damping ratio parameter are read, the unit reads the acceleration, the displacement, the force and the moment of the compressor in the trial run stage, and after the unit is successfully installed and debugged on site, the operation tests of the compressor under the conditions of no load, loading, different air inlet pressures and different discharge capacities are carried out through the data acquisition module, and the operation parameters of each component are recorded.
Preferably, in step S3, the method for establishing the simulation mathematical model specifically includes:
establishing an ideal thermodynamic equation as follows: p is f (n, Q, T1, k, B, r1, r2, P1, P2, z, Q, T), the thermodynamic equations of each stage are combined to form an equation set for calculation after the establishment is finished, a dynamic pressure value P is output for kinetic calculation and vibration mechanics calculation after the calculation is finished, and f (P) is an ideal thermodynamic mathematical model of the unit, and f (P) is output for a neural network;
establishing an ideal kinetic equation as follows: w ═ f (n, m, t, l, P1, P2, r2, c, P, z1), output torque value W is used for vibration mechanics calculation after establishment is finished, and f (W) is output for neural network when f (W) is an ideal dynamics mathematical model;
f, (P) and f (W) are combined with a BP neural network method, real-time operation parameters obtained by a field communication module are input, fitting parameters z and z1 are iteratively optimized, and a compressor simulation thermodynamic model and a simulation kinetic model are established;
establishing an ideal vibration mechanical equation: f (M, C, K, P, W, t), making f (V) be an ideal vibration mechanics mathematical model, combining finite element analysis and real-time operation parameters obtained by a field communication module to modify structure dynamics, making f (V) be a simulation vibration mechanics model modified by SDM, and outputting f (V) for a neural network and an expert system;
the method comprises the following steps that n is the rotating speed of a compressor, Q is the exhaust volume, Q is the clearance volume, T is time, T is the air inlet temperature, T1 is the exhaust temperature, K is the gas adiabatic index, B is the heat exchange value between the unit time of a cylinder and the outside, r1 is the diameter value of the cylinder at a certain level, r2 is the revolution radius of a crankshaft, P1 is the air inlet pressure, P2 is the exhaust pressure, Q is the gas displacement, z is a thermodynamic fitting parameter, P is the transient pressure in the cylinder, M is the reciprocating inertia mass force, l is the number of the rows of the compressor, C is sliding friction damping, z1 is a kinetic fitting parameter, W is the transient torque value output by a motor, M is the mass matrix of each part of the compressor, C is the damping matrix of each part of the compressor, and K is the rigidity matrix of each part of the.
Further, when a simulated vibration mechanics model is established, a compressor model is firstly subjected to discrete decomposition to form each substructure, then a main mode and a constraint mode of each substructure are obtained through finite element simulation, a transformation matrix is multiplied to obtain a polycondensation rigidity matrix and a polycondensation quality matrix of a mode space, structural dynamics modification based on finite element analysis and experimental modes is carried out through comparison with mode data of each substructure of an actually measured compressor, then a quality matrix, a rigidity matrix and a damping matrix of the mode space of the whole machine are obtained through combination, generalized force is introduced to carry out unit coupling analysis, and acceleration, displacement, force and moment of corresponding parts of the compressor are extracted.
Preferably, in step S3, the filtering is kalman filtering.
Compared with the prior art, the invention has the following beneficial technical effects:
1. the invention adopts the neural network algorithm to combine with the real-time measured data of field installation and debugging to carry out fitting optimization on the thermal and dynamic mathematical model of the compressor model, thereby greatly improving the simulation precision of the mathematical model on the compressor. In the fault diagnosis process, a reference standard is provided for accurately judging the fault reason and the fault position.
2. The method adopts a dynamic modification (SDM) mode based on a finite element structure model and measured data to improve the accuracy of the vibration simulation of the compressor, can directly simulate the vibration characteristics of the compressor when various faults occur by combining an accurate thermal and dynamic model and a simulation module based on a neural network, and greatly improves the success rate and the accuracy by comparing the mode of positioning and judging the faults by extracting corresponding fault characteristic spectrums from the time domain and the frequency domain of the vibration characteristics of the unit.
3. The invention can track and detect in the process of production, assembly and debugging (or normal operation) of each set of unit, and extracts the perfect mathematical models (f (P) ', f (W) ', f (V) ') of the measured parameters, so that the mathematical models are only used for the detected unit, and the perfect mathematical models of the compressor have no universality and improve the simulation precision (even if the mathematical models of the compressors in the same model and batch have slight difference, so as to better simulate the operation of the compressor unit).
4. The invention can diagnose the fault in a real-time online monitoring mode, can only reserve a sensor mounting hole on the unit, and can install corresponding detection equipment to diagnose the field fault when the fault symptom occurs (generally found by a PLC monitoring system or manual inspection of a compressor). Compared with other fault diagnosis systems which need to be monitored on line in real time, the fault diagnosis cost is greatly reduced.
Drawings
Fig. 1 is a flow chart of the fault diagnosis of the present invention.
FIG. 2 is a schematic diagram of a thermodynamic model and a kinetic model fitting process.
FIG. 3 is a schematic diagram of a state of a neural network-based parameter fit.
Detailed Description
The present invention will now be described in further detail with reference to specific examples, which are intended to be illustrative, but not limiting, of the invention.
The invention provides a fault diagnosis system of a reciprocating compressor based on a neural network algorithm, which comprises a field communication module, a data acquisition module, a simulation module based on a neural network, an expert system module based on standard fault characteristics, a fault diagnosis upper computer and a host computer, wherein the field communication module is used for acquiring data;
the field communication module is used for acquiring real-time operation parameters of the compressor from the PLC monitoring system; the field communication module supports various communication protocols such as a Modbus protocol, a TCP/IP protocol and a CAN bus.
The data acquisition module is used for acquiring characteristic parameters of each part of the compressor; the data acquisition module comprises temperature, pressure, stress, vibration and eddy current sensors (each sensor is arranged in a reserved sensor mounting hole of each part of the compressor), a data acquisition card and data acquisition software, and is used for acquiring each operation parameter required by thermodynamic, dynamic and vibration mechanical equations of the compressor and attribute parameters (such as n, q, T, T1, k, B, r1, r2, p1, p2, m, v, T, l, p2, r2 and c) of corresponding parts. When the unit is machined, the components such as the air cylinder, the middle body, the machine body, the cylinder head support, the middle body support, the buffer tank and the washing tank are tested through the data acquisition module, the natural frequency, the modal vibration mode and the damping ratio parameter are read, the unit reads the acceleration, the displacement, the force and the moment of the compressor in the commissioning stage, and after the unit is successfully installed and debugged on site, the operation tests of the compressor under the conditions of no load, loading, different air inlet pressures and different discharge capacities are carried out through the data acquisition module, and the operation parameters of each component are recorded.
And the data acquired by the communication module and the data acquisition module are both sent to the simulation module based on the neural network.
The simulation module based on the neural network adopts the design data of the compressor, the characteristic parameters of each part acquired by the data acquisition module during the normal working period of the compressor and the real-time operation parameters acquired by the field communication module, and establishes a simulation thermodynamic, kinetic and vibration mechanical equation after filtering. The filtering mode can adopt Kalman filtering.
The thermodynamic equation is: f (n, Q, T1, k, B, r1, r2, P1, P2, z, Q, T)
The method comprises the following steps of firstly, establishing a thermodynamic equation set, wherein n is the rotating speed of a compressor, Q is the displacement, Q is the clearance volume, T is time, T is the air inlet temperature, T1 is the air outlet temperature, k is the gas adiabatic index, B is the heat exchange value of the cylinder with the outside in unit time, r1 is the radius of the cylinder, r2 is the revolution radius of a crankshaft, P1 is the air inlet pressure, P2 is the air outlet pressure, Q is the air discharge capacity, P is the transient pressure in the cylinder, and z is a thermodynamic fitting parameter (different z values are selected according to different working conditions to enable the equation result P to be close to the transient pressure value P in the cylinder acquired by a field pressure sensor), the thermodynamic equations at all levels are combined to form an equation set for calculation after the completion of the establishment, the gas dynamic pressure P excitation parameter is output for the calculation of thermodynamic vibration mechanics after the. The thermodynamic fitting process is to input real-time working condition parameters and design parameters measured by temperature, pressure, stress, vibration and eddy current sensors installed on the compressor unit and fit the discharge temperature, discharge capacity and discharge pressure of the compressor unit under different working conditions by adjusting the weight and the hidden layer number of the neural network.
The kinetic equation is: w ═ f (n, m, t, l, P1, P2, r2, c, z1, P)
Wherein, W is the transient force of the moving part, n is the rotating speed of the compressor, m is the reciprocating inertia mass force, t is the time, l is the number of the rows of the compressor, c is the sliding friction damping, z1 is the dynamics fitting parameter (according to different working conditions, different z1 values are selected to enable the equation result to be close to the W acquired by the field stress sensor), the W is output for the calculation of the vibration mechanics after the calculation is finished, and an ideal dynamics model f (W) is output for the neural network to use. The dynamic fitting process is to input tested power, rotating speed, input torque parameters and design parameters and fit the stress change of the shafting of the compressor unit under different working conditions by adjusting the weight and the hidden layer number of the neural network.
And inputting real-time operation parameters acquired by a field communication module into an ideal thermodynamic model f (P) and an ideal kinetic mathematical model f (W) by combining a BP neural network method, iteratively optimizing fitting parameters z and z1, wherein f (P) is a correction form of f (P), and f (W) is a correction form of f (W), so that simulation mathematical models (f (P') and f (W)) which can accurately reflect the thermal and kinetic characteristics of the compressor and meet the requirement of engineering precision are established.
The vibration mechanics equation: v ═ f (M, C, K, P, W, t)
Wherein V is a vibration value of the unit, M is a mass matrix, C is a damping matrix, and K is a rigidity matrix.
In the process of vibration mechanics calculation, a compressor model is firstly subjected to discrete decomposition to form each substructure, such as a cylinder, a middle body, a machine body and other substructures, then a main mode and a constraint mode of each substructure are obtained through finite element simulation, a transformation matrix is multiplied to obtain a polycondensation rigidity matrix and a polycondensation quality matrix of a mode space, structural dynamics modification based on finite element analysis and experimental modes is carried out through comparison with mode data of each substructure of an actually measured compressor, then a quality matrix, a rigidity matrix and a damping matrix of the mode space of the whole machine are obtained through combination, a generalized force is introduced to carry out machine set coupling analysis, and vibration acceleration, displacement, force and moment of the corresponding part of the compressor can be extracted.
The method specifically comprises the following steps:
obtaining a mass matrix M under the modal coordinates of each substructure through finite element simulationssStiffness matrix KmmWherein the degree of freedom of the abutting boundary points of the substructures is positioned as the main degree of freedom xmThe degree of freedom of the non-boundary point is defined as the degree of freedom xsAnd X is the degree of freedom of the substructure, the motion of the substructure can be represented by the following equation:
wherein M isms，MsmAnd Kms，KsmCoupling matrices, f, of master and slave degrees of freedom, respectivelymAs a butting force, f0mFor the butt-joint force amplitude, let D be K-omega2M is a power matrix, fm＝f0meiωtWhen there is a response displacement x ═ XeiωtThen, the first formula in the above formula is solved
The degree of freedom of the substructure may be denoted asOrder to
Then TdObtaining a polycondensation quality matrix under the ith substructure main coordinate for transforming the matrix
And polycondensation stiffness matrix
Is composed of
The mass, rigidity and damping matrix of the modal space of the whole compressor are obtained through the combination of the substructures and the modes, the generalized force is introduced to carry out the coupling analysis of the unit, the acceleration, the displacement, the force and the torque of the corresponding part of the compressor can be extracted, and a simulated vibration mechanics model f (V) is output to be used by a neural network and an expert system. Referring to FIG. 3, the XX substructure represents a substructure model of other components such as the mid-body support, inlet and exhaust surge tanks, base, etc.
The expert system module comprises a comprehensive database, an expert knowledge base and an inference machine. The comprehensive database comprises a simulation thermodynamic model, a simulation dynamic model, a simulation vibration mechanical model and the fault characteristic data, wherein the simulation thermodynamic model, the simulation dynamic model and the simulation vibration mechanical model are fitted by a simulation module during the normal working period of the reciprocating compressor. The expert knowledge base contains the judgment criteria of the faults and the processing method aiming at various faults. The inference engine is responsible for applying the control strategy to find the applicable fault judgment criteria. The expert system module can be read, added and modified. The integrated database and the expert knowledge base can be read, added, and modified.
And adjusting the fitting parameters z and z1 of the two mathematical models, and establishing an accurate compressor simulation mathematical model by combining a BP neural network method (see figure 2), and storing the model to an expert system for fault diagnosis and comparison. Then, different simulation mathematical model parameters are changed, fault data characteristics corresponding to various faults at different parts of the compressor are simulated (such as reducing the first-stage exhaust pressure p1, reducing the first-stage cylinder volumetric efficiency, simulating the first-stage exhaust valve air leakage fault, increasing a transient impact force on the reciprocating inertia force m to simulate the piston knocking fault when the piston moves to the top dead center, reducing the K value of a coupler to simulate the bolt loosening fault of the coupler, increasing the C value at the second row of the machine body to simulate the main bearing lubrication fault of the second row of the crankshaft, and the like), and the fault data characteristics are stored in a fault library module based on an expert system, and the fault library module can be read, added and modified.
The fault diagnosis host is used for facing a user, and the fault diagnosis process, the result, the elimination measure and the like can be read and displayed through the interface; the upper computer is used for facing developers, and the simulation module and the expert system module can be read, added and modified through the interface.
As shown in fig. 1, the present invention also provides a fault diagnosis method for a reciprocating compressor based on a neural network algorithm, comprising the steps of:
s1, acquiring characteristic parameters of each part of the compressor through the data acquisition module, and sending the characteristic parameters to the simulation module based on the neural network;
the method specifically comprises the following steps: when a machine set is machined, the data acquisition module is used for testing components such as an air cylinder, a middle body, a machine body, a cylinder head support, a middle body support, a buffer tank, a washing tank and the like, and reading invariable attribute parameters such as natural frequency, modal shape and the like (used for calculating M, C, K, z2 and other parameters by combining design parameters such as length, width, height, density, elastic modulus and the like); the unit reads the acceleration, displacement, force and moment of the compressor in the test run stage; after the unit is successfully installed and debugged on site, the compressor is subjected to operation tests under no-load, loading, different air inlet pressures and different discharge capacities, and parameters (such as n, v (T), q, T, k, B, r1, r2, p1, p2, m, v, T, l, p2, r2 and c) of the compressor, such as thermal, mechanical and vibration characteristics, are detected through a PLC (programmable logic controller) and a data acquisition module; all parameters are sent to the neural network based simulation module.
S2, acquiring real-time operation parameters of the compressor PLC monitoring system through the field communication module, and sending the real-time operation parameters to the simulation module based on the neural network;
s3, simulation: the simulation module based on the neural network carries out simulation: in the compressor assembling and trial run stage, a simulation module of the neural network adopts compressor design data, characteristic parameters of each part of the compressor and real-time operation parameters of the compressor, which are acquired by a data acquisition module, and establishes an ideal thermodynamic equation, an ideal kinetic equation and an ideal vibration mechanical equation of the compressor after filtering and removing noise to obtain an ideal mathematical model of the compressor; in the loading operation stage of the compressor, real-time operation parameters acquired by a field communication module are input into an ideal mathematical model, a simulation thermodynamic model, a simulation kinetic model and a simulation vibration mechanical model of the compressor are established by combining a BP neural network method and structural dynamics modification, the simulation mathematical model of the compressor is obtained, and the established simulation mathematical model is stored in an expert system module; different parameters of relevant fault parts in the simulation mathematical model are changed through the neural network algorithm iteration, various faults of different parts of the compressor are simulated to obtain fault data characteristics, and the fault data characteristics are stored in the expert system module
The method specifically comprises the following steps: when the simulation module based on the neural network calculates the heat and mechanics of the compressor, the existing thermodynamic and kinetic empirical calculation methods f (P), f (W) and parameters z, z1 are corrected by combining the neural network algorithm and field test data, and mathematical models (f (P)', f (W)) capable of accurately reflecting the heat and kinetic characteristics of the compressor are established; when the vibration characteristic of the compressor is calculated, the compressor is firstly split into a plurality of substructures, a finite element model of the substructure of the compressor is established, parameters such as natural frequency, modal vibration mode, damping ratio and the like of the substructure are obtained through actual measurement, structural dynamics correction (SDM) based on experimental test is carried out on the substructure model, the mass, rigidity and damping matrix of the modal space of the whole compressor are obtained through substructure modal combination, generalized force is introduced to carry out unit coupling, and a mathematical model f (V) capable of accurately reflecting the vibration characteristic of the compressor is established.
S4, a fault diagnosis process, wherein when a fault symptom occurs, the real-time operation parameters of the compressor acquired by the field communication module are used as fault data characteristics to be matched, and the fault data characteristics to be matched and the fault data characteristics stored in the expert system module are judged to be matched according to the fault judgment criteria in the expert knowledge base; if the matching is successful, fault diagnosis is realized, and corresponding expert opinions are given; if the matching is unsuccessful, relevant fault part parameters of the compressor simulation mathematical models (f (P) ', f (W) ', f (V) ') are modified iteratively through a neural network algorithm until the simulated fault data characteristics are successfully matched with the fault data characteristics to be matched, then a fault elimination method in an expert knowledge base is called out to complete fault diagnosis, and the simulated fault data characteristics are stored in an expert system module for selection when the same type of fault occurs next time.
The method specifically comprises the following steps: when the compressor operates under a certain working condition, a fault symptom suddenly occurs, for example, the PLC control cabinet of the compressor displays that the exhaust temperature of a certain cylinder is abnormally increased, the power of a unit is reduced, and the vibration of a field detection cylinder head is increased. However, there are many possibilities of causing such an abnormal failure, and it is difficult to directly determine the cause and the location of the failure. At the moment, a sensor of the system can be arranged at a reserved part of the compressor, and the system detects data such as the pressure, the temperature and the pulsation of an air cylinder of the compressor, the horizontal runout of a piston rod, the torsional vibration of a crankshaft, the vibration of a machine body, the air cylinder and a support and the like; and comparing and judging the fault position or fault reason (such as the fault of the exhaust valve position) by combining normal mathematical models (f (P) ', f (W) ', f (V) ') of the compressor under the working condition, and calling an expert system to compare and search the fault reason to finish fault diagnosis. If the fault characteristics do not exist (such as the fault library does not store the gas valve faults), parameters of relevant fault parts of a normal compressor simulation mathematical model can be iteratively modified through a neural network algorithm (such as clearance volume modification in a thermodynamic model (f (P) '), and rigidity and quality parameters (K, M) of an exhaust valve are modified in a vibration mechanics model (f (V)') until fault data characteristics of the on-site compressor are accurately simulated; then, the method for eliminating the fault in the expert knowledge base is called out (such as replacing an exhaust valve spring and a valve plate), fault diagnosis is completed, and the fault data characteristics are stored in the fault expert knowledge base so that the system can quickly judge and complete fault base accumulation when similar faults occur next time.
Claims (5)
1. A fault diagnosis method of a reciprocating compressor based on a neural network algorithm is characterized by comprising the following steps:
s1, acquiring characteristic parameters of each part of the compressor through the data acquisition module, and sending the characteristic parameters to the simulation module based on the neural network;
s2, acquiring real-time operation parameters of the compressor PLC monitoring system through the field communication module, and sending the real-time operation parameters to the simulation module based on the neural network;
s3, carrying out simulation by a simulation module based on the neural network: in the compressor assembling and trial run stage, a simulation module of the neural network adopts compressor design data, characteristic parameters of each part of the compressor and real-time operation parameters of the compressor, which are acquired by a data acquisition module, and establishes an ideal thermodynamic equation, an ideal kinetic equation and an ideal vibration mechanical equation of the compressor after filtering and removing noise to obtain an ideal mathematical model of the compressor; in the loading operation stage of the compressor, real-time operation parameters acquired by a field communication module are input into an ideal mathematical model, a simulation thermodynamic model, a simulation kinetic model and a simulation vibration mechanical model of the compressor are established by combining a BP neural network method and structural dynamics modification, the simulation mathematical model of the compressor is obtained, and the established simulation mathematical model is stored in an expert system module; iteratively changing different parameters of relevant fault parts in the simulation mathematical model through a neural network algorithm, simulating various faults of different parts of the compressor to obtain fault data characteristics, and storing the fault data characteristics to an expert system module;
s4, fault diagnosis: when a fault symptom occurs, the real-time operation parameters of the compressor acquired by the field communication module are used as fault data characteristics to be matched, and the fault data characteristics to be matched and the fault data characteristics stored in the expert system module are judged to be matched according to a fault judgment criterion in the expert knowledge base; if the matching is successful, fault diagnosis is realized, and corresponding expert opinions are given; if the matching is unsuccessful, relevant fault part parameters of the compressor simulation mathematical model are modified through the neural network algorithm in an iterative mode until the simulated fault data characteristics are successfully matched with the fault data characteristics to be matched, then a fault elimination method in an expert knowledge base is called out, fault diagnosis is completed, and the simulated fault data characteristics are stored in an expert system module.
2. The neural network algorithm-based reciprocating compressor fault diagnosis method of claim 1, wherein: step S1 specifically includes: when the unit is machined, the air cylinder, the middle body, the machine body, the cylinder head support, the middle body support, the buffer tank and the washing tank are tested through the data acquisition module, the natural frequency, the modal vibration mode and the damping ratio parameter are read, the unit reads the acceleration, the displacement, the force and the moment of the compressor in the trial run stage, and after the unit is successfully installed and debugged on site, the operation tests of the compressor under the conditions of no load, loading, different air inlet pressures and different discharge capacities are carried out through the data acquisition module, and the operation parameters of each component are recorded.
3. The neural network algorithm-based reciprocating compressor fault diagnosis method of claim 1, wherein: in step S3, the method for establishing the simulation mathematical model specifically includes:
establishing an ideal thermodynamic equation as follows: p is f (n, Q, T1, k, B, r1, r2, P1, P2, z, Q, T), the thermodynamic equations of each stage are combined to form an equation set for calculation after the establishment is finished, a dynamic pressure value P is output for kinetic calculation and vibration mechanics calculation after the calculation is finished, and f (P) is an ideal thermodynamic mathematical model of the unit, and f (P) is output for a neural network;
establishing an ideal kinetic equation as follows: w ═ f (n, m, t, l, P1, P2, r2, c, P, z1), output torque value W is used for vibration mechanics calculation after establishment is finished, and f (W) is output for neural network when f (W) is an ideal dynamics mathematical model;
f, (P) and f (W) are combined with a BP neural network method, real-time operation parameters obtained by a field communication module are input, fitting parameters z and z1 are iteratively optimized, and a compressor simulation thermodynamic model and a simulation kinetic model are established;
establishing an ideal vibration mechanical equation: f (M, C, K, P, W, t), making f (V) be an ideal vibration mechanics mathematical model, combining finite element analysis and real-time operation parameters obtained by a field communication module to modify structure dynamics, making f (V) be a simulation vibration mechanics model modified by SDM, and outputting f (V) for a neural network and an expert system;
the method comprises the following steps that n is the rotating speed of a compressor, Q is the exhaust volume, Q is the clearance volume, T is time, T is the air inlet temperature, T1 is the exhaust temperature, K is the gas adiabatic index, B is the heat exchange value between the unit time of a cylinder and the outside, r1 is the diameter value of the cylinder at a certain level, r2 is the revolution radius of a crankshaft, P1 is the air inlet pressure, P2 is the exhaust pressure, Q is the gas displacement, z is a thermodynamic fitting parameter, P is the transient pressure in the cylinder, M is the reciprocating inertia mass force, l is the number of the rows of the compressor, C is sliding friction damping, z1 is a kinetic fitting parameter, W is the transient torque value output by a motor, M is the mass matrix of each part of the compressor, C is the damping matrix of each part of the compressor, and K is the rigidity matrix of each part of the.
4. The neural network algorithm-based reciprocating compressor fault diagnosis method of claim 3, wherein: when a simulation vibration mechanics model is established, a compressor model is firstly subjected to discrete decomposition to form each substructure, then a main mode and a constraint mode of each substructure are obtained through finite element simulation, a transformation matrix is multiplied to obtain a polycondensation rigidity matrix and a polycondensation quality matrix of a mode space, structural dynamics modification based on finite element analysis and experimental modes is carried out through comparison with mode data of each substructure of an actually measured compressor, then a quality matrix, a rigidity matrix and a damping matrix of a mode space of the whole machine are obtained through combination, generalized force is introduced to carry out unit coupling analysis, and acceleration, displacement, force and moment of corresponding parts of the compressor are extracted.
5. The neural network algorithm-based reciprocating compressor fault diagnosis method of claim 1, wherein: in step S3, kalman filtering is used as the filtering.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811643547.8A CN109779894B (en) | 2018-12-29 | 2018-12-29 | Reciprocating compressor fault diagnosis system and method based on neural network algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811643547.8A CN109779894B (en) | 2018-12-29 | 2018-12-29 | Reciprocating compressor fault diagnosis system and method based on neural network algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109779894A CN109779894A (en) | 2019-05-21 |
| CN109779894B true CN109779894B (en) | 2020-01-10 |
Family
ID=66499520
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811643547.8A Active CN109779894B (en) | 2018-12-29 | 2018-12-29 | Reciprocating compressor fault diagnosis system and method based on neural network algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109779894B (en) |
Families Citing this family (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111980902B (en) * | 2019-05-23 | 2022-08-12 | 睿捷国际股份有限公司 | Monitoring system and monitoring method for reciprocating compressor |
| CN112051039A (en) * | 2019-06-05 | 2020-12-08 | 大连中和聚能自动控制***有限公司 | Experiment table system for fault simulation of pneumatic actuating mechanism |
| CN112049836B (en) * | 2019-06-05 | 2023-06-02 | 大连中和聚能自动控制***有限公司 | Visual system of pneumatic actuating mechanism |
| CN110360091B (en) * | 2019-06-05 | 2020-08-28 | 合肥通用机械研究院有限公司 | Refrigeration compressor measurement and control system and method based on neural network |
| CN110596584A (en) * | 2019-07-12 | 2019-12-20 | 贵州航天林泉电机有限公司 | Test bed system for high-speed generator |
| CN110318993B (en) * | 2019-08-01 | 2020-09-22 | 江西资生科技有限公司 | Reciprocating compressor connecting rod slideway remote fault diagnosis system and device |
| CN112443479A (en) * | 2019-08-27 | 2021-03-05 | 深圳中集智能科技有限公司 | Compressor fault diagnosis method and device, storage medium and electronic equipment |
| EP3786746A1 (en) | 2019-08-28 | 2021-03-03 | Siemens Aktiengesellschaft | Efficient error analysis by means of simulated errors in a digital twin |
| CN111027260A (en) * | 2019-12-24 | 2020-04-17 | 大连圣力来监测技术有限公司 | Reciprocating compressor fault diagnosis method based on neural network |
| CN111275007B (en) * | 2020-02-24 | 2021-10-22 | 山东大学 | Bearing fault diagnosis method and system based on multi-scale information fusion |
| CN111272412B (en) * | 2020-03-23 | 2022-02-22 | 北京昊鹏智能技术有限公司 | Fault detection method, device, equipment and storage medium of electric control pneumatic equipment |
| CN112926257A (en) * | 2020-09-25 | 2021-06-08 | 中国石油天然气集团有限公司 | Reciprocating natural gas compressor fault diagnosis system and diagnosis method |
| CN112610420B (en) * | 2020-12-17 | 2024-04-02 | 华能辽宁清洁能源有限责任公司 | Novel wind generating set safety chain simulation maintenance method |
| CN113011055B (en) * | 2021-02-03 | 2023-01-03 | 中国船级社 | Motor high-temperature fault simulation method and system based on finite element technology |
| CN112945535B (en) * | 2021-02-20 | 2023-04-18 | 广东石油化工学院 | Rotating machinery fault detection method and device based on numerical simulation |
| CN113435111B (en) * | 2021-06-08 | 2022-10-25 | 西安交通大学 | Fault diagnosis method and system for reciprocating compressor |
| CN113656989A (en) * | 2021-06-29 | 2021-11-16 | 东方电气集团东方电机有限公司 | Bolt fault diagnosis method and device |
| CN113494527B (en) * | 2021-07-30 | 2022-06-24 | 哈尔滨工业大学 | Constant force control method based on electromagnetic auxiliary type constant force spring support |
| CN113587363B (en) * | 2021-08-04 | 2022-07-26 | 广东美的暖通设备有限公司 | Compressor fault detection method and device, computing equipment and storage medium |
| CN113757093B (en) * | 2021-10-14 | 2023-06-16 | 中国海洋石油集团有限公司 | Flash steam compressor unit fault diagnosis method |
| CN114184883A (en) * | 2021-11-22 | 2022-03-15 | 国网河南省电力公司漯河供电公司 | Distribution network fault detection precision calculation method based on distribution network fault simulation |
| CN115356096B (en) * | 2022-08-18 | 2024-05-14 | 西安交通大学 | System and method for researching surge characteristics of compressor pipe network system |
| CN115596654B (en) * | 2022-09-21 | 2023-12-22 | 西安交通大学 | Reciprocating compressor fault diagnosis method and system based on state parameter learning |
| CN115640915B (en) | 2022-12-19 | 2023-03-10 | 成都秦川物联网科技股份有限公司 | Intelligent gas pipe network compressor safety management method and Internet of things system |
| CN116538092B (en) * | 2023-07-06 | 2023-11-14 | 中国科学院理化技术研究所 | Compressor on-line monitoring and diagnosing method, device, equipment and storage medium |
| CN117370824B (en) * | 2023-12-07 | 2024-03-22 | 深圳市伟昊净化设备有限公司 | Intelligent monitoring method and system for air inlet state of air compressor |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5919267A (en) * | 1997-04-09 | 1999-07-06 | Mcdonnell Douglas Corporation | Neural network fault diagnostics systems and related method |
| CN104750742A (en) * | 2013-12-31 | 2015-07-01 | 南京理工大学常熟研究院有限公司 | Fault diagnosis method and system for heading machine hydraulic system |
| CN106124982A (en) * | 2016-06-14 | 2016-11-16 | 都城绿色能源有限公司 | Automatic expert's resultant fault diagnostic system of a kind of Wind turbines and diagnostic method |
-
2018
- 2018-12-29 CN CN201811643547.8A patent/CN109779894B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN109779894A (en) | 2019-05-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109779894B (en) | Reciprocating compressor fault diagnosis system and method based on neural network algorithm | |
| JP4589751B2 (en) | Method and apparatus for determining the life of a turbocharger | |
| CN100555238C (en) | Be used for the emergency protection of vacuum pump and the accurate diagnostic method and the accurate diagnostic system of anticipatory maintenance | |
| Elhaj et al. | Numerical simulation and experimental study of a two-stage reciprocating compressor for condition monitoring | |
| KR0185796B1 (en) | Design verification device | |
| WO2020019681A1 (en) | Fault diagnosis method and apparatus for oil production equipment | |
| CN106404403B (en) | Method and system for analysis of a turbomachine | |
| JP5897824B2 (en) | Turbomachine risk analysis system and machine-readable medium for storing machine-readable instructions for causing a computer to create an inspection recommendation for the turbomachine | |
| CN101652649A (en) | Crack progress predicting method and program | |
| CN102778360B (en) | Use the state estimation of equivalent time sampling, diagnosis and control | |
| JP2004523735A (en) | Inference signal generator for instrumented equipment and processes | |
| US8903692B2 (en) | Method for the detection of failures in a turbomachine by means of a theoretical model of the thermodynamic cycle of the said turbomachine | |
| CN111089729A (en) | Method and system for forecasting maintenance of a component by means of a solid-state acoustic sensor | |
| WO2018044507A1 (en) | Machine learning-aided model-based method for estimating gas turbine component life | |
| CN113420465B (en) | Hydraulic support life cycle management method based on digital twin model | |
| JP3351925B2 (en) | Equipment management system | |
| CN114912371A (en) | Digital twin body modeling system and method for running stability detection of tubular pump | |
| CN114154360B (en) | Multi-physical field probability digital twin modeling method for high-pressure internal gear pump | |
| CN115596588A (en) | Online fault diagnosis equipment and diagnosis method for oil injector | |
| CN116090266B (en) | Real-time fault diagnosis method for air path performance of power device | |
| KR100905235B1 (en) | A precision diagnostic method for the failure protection and predictive maintenance of a vacuum pump and a precision diagnostic system therefor | |
| CN117309341A (en) | Cylinder liner fatigue testing device and cylinder liner testing method | |
| AU2022201697B2 (en) | Method and system of multi-objective and multi-dimensional online joint monitoring for nuclear turbine | |
| CN108757224B (en) | Fault diagnosis method for impact test of solid rocket engine | |
| CN111428396A (en) | Method for evaluating abrasion degree of hinge pin shaft of online loader |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |