CN105004462A - Fault-identification-based fan energy consumption monitoring system - Google Patents

Abstract

The invention discloses a fan energy consumption monitoring system based on fault identification, which includes three triaxial acceleration sensors respectively placed on the housing of the bearing housing, the housing of the motor and the housing of the fan, and two eddy currents perpendicular to each other in the vertical plane of the rotating shaft The sensor, the triaxial acceleration sensor and the eddy current sensor are connected with the signal processing and feature extraction module, and the signal processing and feature extraction module are connected with the neural network-based classification recognition module. In all-weather energy consumption monitoring, it is only necessary to install a low-cost triaxial acceleration sensor and eddy current sensor on the fan, and input the three-dimensional vibration signal and the axis trajectory feature vector into the multi-weight neural network, and the network output is the classification of energy consumption increase.

Description

Fan energy consumption monitoring system based on fault identification

This application is a divisional application with application number: 201410257423.1, application date: 2014.6.11, and name "Fan Energy Consumption Monitoring Method and System".

technical field

The invention relates to a method and system for monitoring fan energy consumption.

Background technique

Fans, motors, pumps, and compressors are collectively referred to as "industrial motor systems" by the International Energy Agency (IEA). The National Development and Reform Commission pointed out in the "Eleventh Five-Year" energy-saving plan that industrial motor systems are the main power users in China, accounting for more than 50% of the total electricity consumption, of which the electricity consumption of fans accounts for 10.4% of the country's electricity consumption. Therefore, the improvement of fan efficiency is of great significance to saving electric energy.

The fan system has a large volume and a wide area, and the potential for power saving is huge. The promulgation of the national standard "Energy Efficiency Limits and Energy Efficiency Grades of Fans" stipulates the energy efficiency grades, energy efficiency limit values, energy-saving evaluation values and test methods of fans, which provides a basis for the research and development of energy efficiency work of high-efficiency fans in my country. It also marks the beginning of the energy efficiency labeling work for high-efficiency fans in my country. On November 1, 2010, fans will be officially included in the product catalog (sixth batch) of the People's Republic of China to implement energy efficiency labels, and the implementation has begun.

When the fan runs for a long time in industrial production, many failures will occur, such as vibration caused by dynamic unbalance (including the residual unbalance in the manufacturing process of the rotor system; during the rotation of the fan blade, due to local wear or corrosion, and local damage or blockage of foreign matter and other reasons; the blower works under high temperature and high pressure, and the shaft bending phenomenon is caused by thermal deformation and thermal expansion, etc.); vibration caused by misalignment (data show that 30% to 50% of the equipment has misalignment problems. Misalignment can cause radial vibration, and axial vibration; it will cause vibration near the bearing of the coupling, and it will also cause vibration at the free end away from the coupling); mechanical looseness (looseness may cause other failures of the machine or may be caused by other failures Caused by wear and deformation of mechanical parts, misalignment of shafting, imbalance, etc. and looseness); vibration caused by oil film oscillation; vibration caused by gas impact; vibration caused by gas pressure fluctuations; vibration caused by harmonic components; Various failures of the fan drive motor, in addition, such as failures of shafts, belt chains, gears, bearings and other transmission mechanisms, poor heat dissipation caused by motor dust condensation, poor lubrication caused by excessive running time or dirt and water pollution, etc. Causes reduced efficiency and increased energy consumption. These failures will generally cause the motor and fan system to generate heat and increase various losses, thereby reducing system efficiency and increasing system energy consumption. Therefore, there is a causal relationship between the existence of various faults and the reduction of energy efficiency (increase in energy consumption). Mining the numerical relationship between various fault characteristics and the reduction in energy efficiency (increase in energy consumption) can be used as a tool for energy consumption monitoring. in accordance with. It is of great significance for enterprise energy efficiency management, timely elimination and replacement of high energy consumption equipment, and targeted maintenance of high energy consumption equipment.

The overall efficiency of a fan is defined as the ratio of the sum of the kinetic and static pressure energy delivered to the gas by the fan to the energy delivered by the motor. It is now used in the fan energy consumption detection system of other institutions such as quality inspection departments, using GB/T 1236-2000 "Industrial Ventilators Use Standardized Air Ducts for Performance Tests" and GB-19761-2009 "Energy Efficiency Limits and Energy Efficiency Grades for Ventilators 》The test method of the ventilator in the national standard builds a test system, which needs to measure multiple parameters such as speed, pressure difference, flow, power, temperature, torque, etc., and the built system is expensive. Moreover, the existing fan energy consumption detection equipment needs to be equipped with additional structures such as air ducts for different types of fans to facilitate flow and wind pressure measurement, and various sensors such as differential pressure sensors, torque sensors, and rotational speed sensors need to be installed. After the fan has been in operation for a long time, various failures will cause increased energy consumption. Therefore, the monitoring of fan energy consumption is conducive to energy efficiency management of enterprises, timely elimination and replacement of high-energy consumption equipment, and targeted maintenance of high-energy consumption equipment. However, the above energy consumption detection methods make the existing equipment unsuitable for energy consumption monitoring of wind turbine application enterprises, and the existing expensive equipment is even more unsuitable for matching and realizing all-weather energy efficiency monitoring for each wind turbine.

With the advancement of the basic national policy of energy saving and emission reduction, the efficiency reduction and energy consumption increase caused by various failures of fans need to attract the attention of all fan application enterprises. In addition, the energy efficiency reduction and energy consumption increase caused by changes in external grid parameters can also be solved The vibration signal is analyzed. On the premise that the equipment has high cost performance, each high-power fan is equipped with an energy consumption monitoring device to realize long-term real-time monitoring of the energy efficiency of the fan and accurately discover the phenomenon of efficiency reduction caused by various mechanical failures, electrical failures or changes in power grid parameters. Enterprise energy efficiency management, timely elimination and replacement of high-energy-consuming equipment, and targeted maintenance of high-energy-consuming equipment are of great significance.

Contents of the invention

The purpose of the present invention is to provide a fan energy consumption monitoring method and system that is beneficial to enterprise energy efficiency management, timely eliminates and replaces high energy consumption equipment, and realizes targeted troubleshooting of high energy consumption equipment.

Technical solution of the present invention is:

A fan energy consumption monitoring method is characterized in that: comprising the following steps:

(1) First collect offline training samples:

(1) Construction of offline training sample collection system

Construct an energy consumption test system, using various sensors including torque sensors, differential pressure sensors, and rotational speed sensors to detect fan flow, fan total pressure, fan static pressure, volume flow, fan shaft power, and finally obtain fan efficiency. Know the energy consumption of the fan;

Three three-axis acceleration sensors are used, which are respectively placed on the housing of the bearing housing, the housing of the motor, and the housing of the fan to obtain the orthogonal vibration signals of the X, Y, and Z axes of the three test points; The eddy current sensors perpendicular to each other collect vibration signals at the same time, and the collected data are respectively used as a graph fitted by horizontal and vertical coordinates, which is the axis trajectory;

(2) The training samples are obtained offline when the fan is not faulty

The "signal processing and feature extraction module" is used for feature extraction, and after multiple measurements, multiple groups of feature samples when there is no fault are obtained; the efficiency value corresponding to the feature sample when there is no fault is positioned as "low energy consumption";

(3) The training samples are obtained offline when the fan is faulty

A variety of faults and combinations of faults are artificially created. Three three-axis acceleration sensors are used to detect the three-dimensional vibration signals of the three-point fan bearing shell, motor shell, and fan shell. Double eddy current sensors are used to detect the axis track signal. " Signal processing and feature extraction module" to perform feature extraction, perform multiple measurements on each type of fault, and obtain feature samples under each type of fault; compare the efficiency values under different fault conditions with the efficiency values without faults, according to The difference is divided into four types from large to small, which are respectively defined as "high energy consumption", "high energy consumption", "medium energy consumption", and "low energy consumption";

(2) Online energy consumption monitoring

Three three-axis acceleration sensors are used to detect the three-dimensional vibration signals of the fan bearing housing, the motor housing, and the fan housing, and the dual eddy current sensors are used to detect the axis trajectory signal, and the "signal processing and feature extraction module" is used to extract the signal features , to obtain the tested samples; the multi-weight neural network is used as the core algorithm of the "neural network-based classification and recognition module", and the training samples obtained by the "offline acquisition module of training samples for energy consumption detection" are used to construct a multi-freedom model in a high-dimensional space. Degree neural network, after completing the construction of the multi-weight neuron network, the five categories of "high energy consumption", "high energy consumption", "medium energy consumption", "low energy consumption" and "low energy consumption" were obtained. Characterize the multi-weight neuron coverage area of different energy consumption levels; calculate the Euclidean distance between the sample to be identified and the multi-weight neuron network coverage area that characterizes each type of energy consumption level, and the Euclidean distance with the sample to be identified will be the shortest The energy consumption level of the class is regarded as the energy consumption level of the sample to be identified, and the fan energy consumption level is classified as the output of the multi-weight neural network.

The specific method for offline acquisition of training samples when the fan is not faulty is as follows:

Denoise the time-domain signals output by the three triaxial acceleration sensors when there is no fault, and use quaternion PCA to perform principal component analysis, and obtain the vibration characteristics when there is no fault on the premise of maintaining the correlation of the three-axis output signals vector;

When the fan rotor is in normal operation without failure, the double eddy current sensor is used to extract the axis trajectory. When there is no failure, the time domain waveform of the vibration signal of the eddy current sensor is a sinusoidal curve, and the two mutually perpendicular sinusoidal signals are synthesized to obtain a circle or ellipse. , extract the geometric size feature, or gray histogram feature, or texture feature of the axis trajectory image as a feature parameter, and combine it with the vibration feature vector obtained by the acceleration sensor to obtain a sample when there is no fault; use the above method to perform multiple tests , to obtain multiple groups of samples when the energy consumption is low.

A fan energy consumption monitoring system is characterized in that it includes three triaxial acceleration sensors respectively placed on the housing of the bearing housing, the motor housing, and the fan housing, and two eddy current sensors perpendicular to each other in the vertical plane of the rotating shaft. The acceleration sensor and the eddy current sensor are connected with the signal processing and feature extraction module, and the signal processing and feature extraction module are connected with the neural network-based classification recognition module.

The invention proposes a fan energy consumption monitoring method based on the analysis of the vibration signal and the axis track signal. This method does not require the installation of fan ducts and other structures in the application. Three triaxial acceleration sensors are used to detect the multi-point three-dimensional vibration signals of the fan (detecting the vibration signals of the motor shell, the bearing seat shell, and the fan shell), and two eddy current The sensor detects the displacement caused by the eccentricity of the main shaft of the fan, and obtains the trajectory of the shaft center. Based on a large number of experiments, the relationship between different types of faults of fans and the increase of fan energy consumption is obtained, and the level of energy consumption increase caused by different faults is classified and identified through a multi-weight neural network.

Looking at the existing domestic fan energy consumption monitoring methods and systems, no unit has yet achieved the design goal proposed by the present invention.

The present invention is to provide a new type of energy consumption monitoring system, which can realize the online identification and monitoring of the increase in fan energy consumption by only using a triaxial acceleration sensor and an eddy current sensor, and avoids the high cost and difficult installation of the existing energy consumption detection system It is beneficial for enterprises to realize the daily energy consumption monitoring of fans, thereby benefiting energy efficiency management of enterprises, timely elimination and replacement of high energy consumption equipment, and targeted troubleshooting of high energy consumption equipment.

Description of drawings

The present invention will be further described below in conjunction with drawings and embodiments.

Fig. 1 is a structural diagram of the energy consumption monitoring system of the present invention. Among them are three-axis acceleration sensor, eddy current sensor, offline acquisition module of training samples for energy consumption detection, signal processing and feature extraction module, and classification and recognition module based on neural network. The "signal processing and feature extraction module" is implemented by software, including denoising, quaternion PCA feature extraction, and axis trajectory feature extraction (geometric size feature, or gray histogram feature, or texture feature). "Neural network-based classification recognition module" realizes energy consumption increase level classification by multi-weight neural network.

Fig. 2 is a schematic diagram of the offline acquisition method of neural network training samples in the present invention.

Fig. 3 is a schematic diagram of the experimental method for obtaining training samples of the neural network.

Figure 4 is a schematic layout of the sensor. The centerlines of the two eddy current sensors intersect with the central axis of the main shaft of the fan, the central lines of the two eddy current sensors are both perpendicular to the central line of the main shaft of the fan, and the central lines of the two eddy current sensors are perpendicular to each other. The three acceleration sensors are respectively fixedly installed on the surface of the motor, the housing of the bearing seat and the housing of the fan.

Fig. 5 is a schematic diagram of the multi-weight neural network recognition process.

Detailed ways

A fan energy consumption monitoring system, including three three-axis acceleration sensors 1 respectively placed on the housing of the bearing seat, the motor casing, and the fan casing, and two eddy current sensors 2 perpendicular to each other in the vertical plane of the rotating shaft, the three-axis acceleration sensor , the eddy current sensor is connected with the signal processing and feature extraction module, and the signal processing and feature extraction module is connected with the neural network-based classification recognition module. "Signal processing and feature extraction module" and "neural network-based classification recognition module" rely on PC or high-performance controller (such as FPGA, etc.) Heart trajectory feature extraction, pattern recognition based on multi-weight neural network.

1. First, collect offline training samples. The specific implementation process is as follows:

The system is only used for offline collection of neural network training samples. After the collection is completed, the system is no longer used in daily energy efficiency monitoring.

1.1 Construction of offline training sample collection system

According to GB/T 1236-2000 "Performance Test of Standardized Air Duct for Industrial Ventilators" and GB-19761-2009 "Energy Efficiency Limits and Energy Efficiency Grades of Ventilators" national standard, build an energy consumption test system, using torque sensors, differential pressure Sensors, speed sensors and other sensors detect fan flow, fan total pressure, fan static pressure, volume flow, fan shaft power, and finally get fan efficiency, so as to know the fan energy consumption.

According to the layout shown in Figure 4, three three-axis acceleration sensors are used, which are respectively placed on the housing of the bearing housing, the housing of the motor, and the housing of the fan to obtain the orthogonal vibration signals of the X, Y, and Z axes of the three test points. Vibration signals are collected simultaneously by two eddy current sensors perpendicular to each other in the plane perpendicular to the rotating shaft, and the collected data are respectively used as horizontal and vertical coordinates to fit the graph, which is the axis trajectory.

1.2 Offline acquisition of training samples when there is no fault in the fan

Using GB/T 1236-2000 "Performance Test of Standardized Air Duct for Industrial Ventilators" and GB-19761-2009 "Energy Efficiency Limits and Energy Efficiency Grades of Ventilators" national standard method to detect the efficiency of ventilators, and obtain the efficiency value when there is no fault , define this state energy consumption as "low energy consumption". Denoise the time-domain signals output by the three triaxial acceleration sensors when there is no fault, and use quaternion PCA to perform principal component analysis, and obtain the vibration characteristics when there is no fault on the premise of maintaining the correlation of the three-axis output signals vector.

When the fan rotor is in normal operation without failure, the double eddy current sensor is used to extract the axis trajectory. When there is no failure, the time domain waveform of the vibration signal of the eddy current sensor is a sinusoidal curve, and the two mutually perpendicular sinusoidal signals are synthesized to obtain a circle or ellipse. , extract the geometric dimension features, or gray histogram features, or texture features of the axis trajectory image as feature parameters, and combine them with the vibration feature vectors obtained by the acceleration sensor to obtain samples when there is no fault. Using the above method, multiple tests are carried out to obtain multiple groups of samples at the time of "low energy consumption".

1.3 Obtain training samples offline when the fan is faulty

A variety of faults are artificially set, such as poor heat dissipation, poor lubrication, eccentric shaft, grid voltage drop or frequency instability (especially when new energy such as wind power is supplied), unbalanced impeller rotation, and transmission mechanism failures (such as belt chains, gears, etc.) , bearing and other failures), motor oil film oscillation, etc. When a fault occurs, the frequency domain of the vibration signal will change. The axis trajectory will appear banana diagram, figure 8, inner ring diagram, irregular diagram and so on. Combining the extracted vibration signal features and axis track features can characterize different faults or fault combinations. When acquiring offline training samples, multiple energy efficiency detections and feature extractions are performed on each fault and combination of multiple faults to obtain multiple sets of samples for each fault and combination of multiple faults.

For each type of fault and combination of multiple faults, the method in Section 1.2 is used to establish a fan efficiency detection system, and the efficiency values under different fault conditions are compared with the efficiency values without faults. According to the difference from large to small, the average Divided into four types, respectively defined as "high energy consumption", "high energy consumption", "medium energy consumption", "low energy consumption".

2. On-line energy consumption monitoring, the specific implementation method is as follows:

For online energy consumption monitoring, the "offline acquisition module of training samples for energy consumption detection" used in Sections 1.2 and 1.3 is no longer used. Only three triaxial acceleration sensors and two eddy current sensors are used. The acceleration sensor and the eddy current sensor are still installed according to the method shown in Figure 4.

Use the training samples collected offline in 1.2 and 1.3 to construct a multi-degree-of-freedom neural network in a high-dimensional space. After completing the construction of the multi-weight neuron network, five representations of different performances can be obtained: "high energy consumption", "high energy consumption", "medium energy consumption", "low energy consumption" and "low energy consumption". Consumption-level multi-weight neuron footprints. When the fan is running, the identification algorithm based on the multi-weight neural network shown in Figure 5 is used, and the signals collected by the triaxial acceleration sensor and the eddy current sensor are used as input samples after feature extraction to calculate the samples to be identified and characterize each The Euclidean distance between the multi-weight neuron network coverage areas of the class energy consumption level, the energy consumption level with the shortest Euclidean distance to the sample to be identified is taken as the energy consumption level to which the sample to be identified belongs. And classify the fan energy consumption level as the output of the multi-weight neural network.

Claims (2)

1. A fan energy consumption monitoring system based on fault identification, which is characterized in that it includes three triaxial acceleration sensors respectively placed on the housing of the bearing housing, the motor housing, and the fan housing, and two mutually perpendicular acceleration sensors in the vertical plane of the rotating shaft. The eddy current sensor, the triaxial acceleration sensor, and the eddy current sensor are connected with the signal processing and feature extraction module, and the signal processing and feature extraction module are connected with the classification recognition module based on the neural network; The monitoring method includes the following steps: (1) First collect offline training samples: (1) Construction of offline training sample collection system Construct an energy consumption test system, using a variety of sensors including torque sensors, differential pressure sensors, and speed sensors to detect fan flow, fan total pressure, fan static pressure, volume flow, fan shaft power, and finally obtain fan efficiency, so as to know the fan energy consumption; Three three-axis acceleration sensors are used, which are respectively placed on the housing of the bearing housing, the housing of the motor, and the housing of the fan to obtain the orthogonal vibration signals of the X, Y, and Z axes of the three test points; The vertical eddy current sensor collects vibration signals at the same time, and the collected data are respectively used as horizontal and vertical coordinates to fit the graph, which is the axis trajectory; (2) The training samples are obtained offline when the fan is not faulty The "signal processing and feature extraction module" is used for feature extraction, and after multiple measurements, multiple groups of feature samples when there is no fault are obtained; the efficiency value corresponding to the feature sample when there is no fault is positioned as "low energy consumption"; (3) The training samples are obtained offline when the fan is faulty A variety of faults and combinations of faults are artificially created. Three three-axis acceleration sensors are used to detect the three-dimensional vibration signals of the three points of the fan bearing shell, the motor shell, and the fan shell. Double eddy current sensors are used to detect the axis track signal. The "signal Processing and Feature Extraction Module" to perform feature extraction, perform multiple measurements on each type of fault, and obtain feature samples under each type of fault; compare the efficiency values under different fault conditions with the efficiency values when there is no fault, and compare The values are divided into four types from large to small, which are respectively defined as "high energy consumption", "high energy consumption", "medium energy consumption", and "low energy consumption"; (2) Online energy consumption monitoring Three three-axis acceleration sensors are used to detect the three-dimensional vibration signals of the fan bearing housing, the motor casing, and the fan casing, and the double eddy current sensors are used to detect the shaft center trajectory signal, and the "signal processing and feature extraction module" is used to extract the features of the signal. Obtain the tested samples; use the multi-weight neural network as the core algorithm of the "neural network-based classification and identification module", and use the training samples obtained by the "offline acquisition module of training samples for energy consumption detection" to construct multi-degree-of-freedom in high-dimensional space Neural network, after completing the construction of the multi-weight neuron network, obtains five representations of "high energy consumption", "high energy consumption", "medium energy consumption", "low energy consumption" and "low energy consumption" Multi-weight neuron coverage area of different energy consumption levels; calculate the Euclidean distance between the sample to be identified and the multi-weight neuron network coverage area that characterizes each type of energy consumption level, and the Euclidean distance with the sample to be identified is the shortest That type of energy consumption level is regarded as the energy consumption level of the sample to be identified, and the fan energy consumption level is classified as the output of the multi-weight neural network. 2. The fan energy consumption monitoring system based on fault identification according to claim 1, characterized in that: the specific method for offline acquisition of training samples when the fan has no faults is: Denoise the time-domain signals output by the three triaxial acceleration sensors when there is no fault, and use quaternion PCA to perform principal component analysis, and obtain the vibration characteristics when there is no fault on the premise of maintaining the correlation of the three-axis output signals vector; When the fan rotor is in normal operation without failure, the double eddy current sensor is used to extract the axis trajectory. When there is no failure, the time domain waveform of the vibration signal of the eddy current sensor is a sinusoidal curve, and the two mutually perpendicular sinusoidal signals are synthesized to obtain a circle or ellipse. , extract the geometric size feature, or gray histogram feature, or texture feature of the axis trajectory image as a feature parameter, and combine it with the vibration feature vector obtained by the acceleration sensor to obtain a sample when there is no fault; use the above method to perform multiple tests , to obtain multiple groups of samples when the energy consumption is low.