JPH10133740A - Method for detecting abnormality of fan and pump for air-conditioning by acoustic method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To exactly detect the various abnormality of a fan and pump for air- conditioning, and to prevent a stop due to a fault by removing the characteristics of a sound pressure signal in a normal time from a sound pressure signal in an abnor mal time. SOLUTION: When comparing a preliminarily measured sound pressure signal Ss in a normal time with a sound pressure signal Sf at the time of measurement and detecting an abnormal signal, a signal is separated into a low frequency area corresponding to a rotational frequency and a high frequency area corresponding to the specific number of oscillation of a member, and the abnormality of a fan and a pump is detected according to whether or not a value obtained by removing the characteristics of the sound pressure signal Ss in a normal time from the sound pressure signal Sf at the time of measurement is beyond a threshold value by using a filter in an AR model to which a linear predicting method is applied. Even when the abnormal signal is very weak, only the abnormal signal components buried in the normal signal can be extracted, and a statistical processing can be executed by using the filter in the AR model. Thus, a non-contact type equipment diagnostic method using sound can be obtained.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a technology for diagnosing equipment by an acoustic method, and more particularly to a method for diagnosing an abnormality of an air conditioning fan or a pump installed in a building. The present invention can be applied to fans and pumps other than those for air conditioning.

[0002]

2. Description of the Related Art Air conditioning equipment is installed in a heat source machine room and an air conditioner room. In large-scale buildings, air-conditioning machine rooms are distributed and installed in various places, and many buildings have dozens to hundreds or more air-conditioning machine rooms. Conventionally, maintenance personnel have visited the air conditioner installed in the machine room to listen to the sound with ears or visually check the vibration condition to diagnose the condition of the equipment. Such a determination requires a high level of skill, and it is difficult to secure enough personnel to reliably detect abnormalities in many air conditioners. From the above background, there is a demand for the introduction of predictive maintenance by monitoring the state of equipment in order to automate and save labor in air conditioning equipment.

[0003] For a rotating device occupying a large number of air conditioners, a condition monitoring by a vibration method is known. However, since the vibration method is a contact type, it is necessary to attach a sensor for each abnormal vibration generation source of each device, and more sensors than the number of devices are required. In addition, it is difficult to install on small parts or built-in components. Therefore, the number of equipment that can be diagnosed and the causes of abnormalities that can be detected are limited. On the other hand, since the acoustic method is a non-contact type, it is not always necessary to provide a sensor for each device, and only a small number of sensors are installed in an air-conditioning machine room where many air-conditioning devices are installed. There is an advantage that the state of the device can be grasped.

As a method of detecting an abnormality by an acoustic method, for example, Japanese Patent Application Laid-Open No. 5-99475, entitled "Noise Diagnosis Apparatus for Air Conditioner", detects a failure from an abnormal noise of an outdoor fan and a compressor. However, since this device measures the sound pressure level and compares it with the nominal value, if the sound signal generated by the abnormality is weak, it can be buried in the sound signal of the entire device and detect the abnormal signal. Can not.

In Japanese Unexamined Patent Application Publication No. 7-43259, "Abnormality Detection Method and Apparatus", an inverse filter is constructed on the basis of an acoustic waveform obtained from an apparatus in a normal state, and an acoustic waveform obtained from an apparatus to be detected. To obtain a residual signal from which signal components unnecessary for analysis have been removed, and analyze the residual signal to detect an abnormality of the apparatus. However, in this method, since the reverse filter is configured without preprocessing the sound pressure signal, the resolution of the reverse filter is low, especially when abnormal noise occurs in a low frequency range such as imbalance of a rotating body. Cannot detect an abnormality. Furthermore, in the case of an abnormal sound in which the sound pressure in a specific band decreases, such as a decrease in the flow rate of a fluid machine, the residual signal also becomes relatively small, and the abnormality cannot be detected by this method.

[0006]

SUMMARY OF THE INVENTION It is an object of the present invention to accurately detect various abnormalities of air conditioning fans and pumps and to measure sound at predetermined time intervals in order to prevent stoppage due to a failure. It is another object of the present invention to provide a diagnostic method for detecting a change in sound pressure signal from a normal state and detecting occurrence of an abnormality.

[0007]

According to the first aspect of the present invention, there is provided a method for detecting sound of an air conditioning fan and a pump and detecting an abnormality of the equipment by measuring sound in advance. When detecting abnormal signals by comparing the sound pressure signal during normal operation with the sound pressure signal during measurement, the signal is divided into a low frequency range corresponding to the rotation frequency and a high frequency range corresponding to the natural frequency of the member. After the separation, the fan and the pump are determined by using a filter based on an AR model to which a linear prediction method is applied, and determining whether a value obtained by removing a characteristic of a sound pressure signal in a normal state from a sound pressure signal in a measurement exceeds a predetermined threshold. To provide a method for detecting abnormalities in a vehicle.

In a second aspect of the present invention, the ratio of the predicted residual energy is determined from both directions using the prediction coefficients of the normal signal and the abnormal signal, and the change in the distance scale δa defined by the sum is detected. Detects air conditioning fan and pump abnormalities.

According to a third aspect of the present invention, a ratio of a prediction residual energy having an amplitude of three times or more the standard deviation of a prediction residual from both directions is obtained by using prediction coefficients of a normal signal and an abnormal signal. An abnormality in the air conditioning fan and the pump is detected by detecting a change in the distance scale δp defined by the sum.

A filter based on an AR model models a characteristic of a population of normal signals and filters only components of the model, so that it is suitable as preprocessing for signal processing. Also, the linear prediction method itself is a well-known method of statistical signal processing, and is effective in speech analysis and the like. The present invention is characterized in that after separating a signal into a low frequency range and a high frequency range, an AR model to which this linear prediction method is applied is used for abnormality diagnosis of a fan and a pump for air conditioning. Even if the abnormal signal is weak, it is possible to extract only the abnormal signal component buried in the normal signal and perform statistical processing using the AR model filter. Will be provided.

In the following description, the AR model, Levins
Terms such as the on-Durbin algorithm and the AIC standard are well-known terms that are described in detail, for example, in "Signal Processing" published by the Society of Instrument and Control Engineers 1982. Hereinafter, preferred embodiments according to the present invention will be described with reference to the accompanying drawings.

[0012]

DESCRIPTION OF THE PREFERRED EMBODIMENTS In order to carry out an acoustic equipment diagnosis method according to the present invention, a fan and a pump are installed in a laboratory consisting of a simple semi-anechoic chamber as shown in FIGS. 1 and 2 (both are plan views). installed. In FIG. 1, a fan 10, a silencer 12, and a damper 14 are arranged in a room, and sound emitted by the fan is collected by a microphone. Juan used one-sided suction sirocco fan.

In FIG. 2, the pump 20, the radiator 22, the expansion tank 24, the globe valves 25 to 30, the pressure gauge, the thermometer, the flow meter, the butterfly valve, the gate valve and the like are connected by piping, and the sound generated by the pump is generated. Was collected with a microphone. Ball valve 28 ~
Numeral 30 serves as a secondary load corresponding to the fan coil unit. The pump used was a single-suction volute pump.

The fan is 20c from the body at the midpoint of both pulleys.
At a position of m, the pump measured a sound pressure signal of a normal state and an artificially generated abnormal state described later at a position of 50 cm immediately above the bearing case. FIG. 3 shows a measurement analysis system.

The sound pressure signal of the device to be detected was measured using a precision sound level meter, and the sound pressure signal was recorded on a digital data recorder. Next, the recorded sound pressure signal is reproduced and separated into a low frequency range of 500 Hz or less corresponding to the rotation frequency and a high frequency range of 5 kHz or more corresponding to the natural frequency of the member, and then recorded again to the digital data recorder. did. Further, the sound pressure signals separated into each frequency range were taken into a personal computer via a data transfer unit, and the subsequent analysis was performed.

The abnormalities of the fan to be detected by the method of the present invention are as follows: (1) Belt tension (300 to 4500 g) (2) Belt flaw (engraved with a cutter or micro grinder) (3) Air volume (0 to 10,000 CMH) (4) Motor reversal (5) Unbalance (addition of weight) (6) Misalignment (movement of pulley position) (7) Bearing damage (engraved with a micro grinder) was intentionally generated and sound was measured.

[0017] The abnormalities of the pump to be detected by the method of the present invention include: (1) a large change in the flow rate (0 to 1300 liter / min); (2) mixing of air; (3) occurrence of cavitation. Packing abnormality (5) Misalignment (movement of motor mounting position) (6) Sound was measured by intentionally generating bearing damage (carved with a micro grinder).

When an abnormality is detected by the acoustic method, if the sound pressure signal generated by the abnormality is weak, it may be buried in the sound pressure signal of the entire device. Therefore, according to the present invention, a linear prediction method is used as a technique for removing a sound pressure signal in a normal state (hereinafter, referred to as a normal signal) from a sound pressure signal in an abnormal state (hereinafter, referred to as an abnormal signal) and improving sensitivity of abnormality detection. Was applied.

First, the prediction residual ε _ss (n) when the normal signal Ss (n) is modeled by the AR model is expressed by the following equation.

(Equation 9) Here, coefficient _ask is a prediction coefficient, and p is a prediction order.

When this prediction coefficient _ask is applied to the abnormal signal Sf (n), the prediction residual ε _fs (n) is expressed by the following equation.

(Equation 10) This is equivalent to filtering an abnormal signal with characteristics (model) of a normal signal.

The prediction coefficient _ask of the normal signal is the prediction residual ε
_{Since the} sum of squares E _ss (predicted residual energy) of _ss (n) is determined to be the minimum, the square of the predicted residual ε _fs (n) obtained from the abnormal signal whose sound pressure increases due to the abnormality Sum E _fs
Is greater than E _ss . Therefore, determining the ratio [delta] _fs of both the prediction residual energy by the following equation.

[Equation 11]

Incidentally, this δ _fs has no symmetry.
That is, even if the prediction residual energy ratio is similarly obtained from the prediction coefficient a _{fk of} the abnormal signal Sf (n), it does not become the same value as δ _fs . In the case of an abnormal signal in which the sound pressure in a specific band decreases, the predicted residual energy E _fs becomes smaller than E _ss . Accordingly, δ _sf is obtained from the prediction residual energy E _ff and E _sf obtained by predicting the abnormal signal with the prediction coefficient a _fk by the following equation.

(Equation 12)

Then, a symmetric distance scale δa is expressed as
δa = δ _fs + δ _sf is defined. FIG. 4 shows these analysis flows. In addition, in order to improve the detection sensitivity of an abnormal signal that emits an impulsive sound such as a bearing flaw, a standard deviation of each prediction residual is obtained, and a distance scale δp having an amplitude three times or more thereof is calculated by the following equation. Defined.

(Equation 13)

Since an infinite length signal cannot be handled in the analysis of the distance scale, a sound pressure signal of 2 seconds in a low frequency range and 0.75 seconds in a high frequency range is analyzed in consideration of the stability of the analysis. And Here, the prediction coefficient of the AR model was solved using the Levinson-Durbin algorithm, and the prediction order was determined based on the AIC standard. In the present invention, when anomaly detection is performed using the distance scales δa and δp, it is found that the index has a sufficient correlation with the degree of anomaly as described later.

In the analysis, in order to confirm the effect of the distance scale δa, a comparison was made with the distance scale δta directly using the sound pressure signal. The distance scale δta is defined as follows.

[Equation 14]

Further, in order to confirm the effect of the distance scale δp, a comparison was made with a distance scale δtp using a sound pressure signal directly. The distance scale δtp is defined as follows.

(Equation 15)

[0027]

[Analysis results] Typical fan belt tension (normal: 12
00g), air volume (normal: 8300CMH), unbalance and pump flow rate (normal: 800 liters / mi)
n), the misalignment, and the analysis results of changing the packing tightening pressure until an abnormal state is shown in FIGS. The vertical axis represents the distance scale, and the horizontal axis represents the changed amount.

FIGS. 5A to 5C show the analysis results of the distance scale δa in the low frequency range by the fan, FIG. 6A shows the analysis results of the distance scale δa in the high frequency range using the pump, and FIGS. 6B to 6C show the results of the analysis using the pump. It is an analysis result of the distance scale δp in a high frequency range.

In the figure, in addition to the distance scales δa and δp defined as the analysis method in the low frequency range, AR
Distance scales δta and δtp directly obtained from the sound pressure signal without filtering by the model are also displayed.

As a result of analysis, it is generally difficult to detect an abnormality at δta, whereas a threshold value at δa is, for example, 2.1.
Abnormality can be detected by setting the value to about 3.0, and the analysis result is stable. Further, the detection sensitivity of δp is higher than that of δtp. For abnormalities other than those shown in the figure, almost all abnormalities can be detected by using the distance scales δa and δp in the low frequency range and the high frequency range, respectively. The result was able to be obtained. This indicates that the method according to the present invention can analyze a sound pressure signal generated due to an abnormality with high sensitivity.

The following has been found out from the above analysis results. (1) The resolution of the AR model is improved by separating the signal into a low frequency region and a high frequency region as preprocessing for analysis. (2) By defining a symmetric distance scale using the prediction coefficients of both the normal signal and the abnormal signal, it is possible to set the threshold value and reliably detect the abnormality. Also, it is possible to detect an abnormality such as a decrease in sound pressure in a specific band. (3) For a shock signal, the distance scale δp has better detection capability than the distance scale δa. (4) In conclusion, precursory phenomena leading to major failures of the fan and pump are sufficiently detectable by sound

[0032]

As described above in detail, according to the detection method of the present invention, the signal is separated into a low frequency range corresponding to the rotational frequency and a high frequency range corresponding to the natural frequency of the member, and then the signal is linearized. By using a filter based on the AR model to which the prediction method is applied, even if the abnormal signal is weak, it is possible to extract only the abnormal signal component buried in the normal signal and perform statistical processing, and use sound. Thus, a non-contact type facility diagnosis method is provided. Since the diagnostic method according to the present invention can be applied to an existing building, the technical effects thereof are extremely remarkable, such as the ability to improve regular maintenance.

[Brief description of the drawings]

FIG. 1 is a plan view showing a state in which an experimental fan is arranged in a room.

FIG. 2 is a plan view showing a state where an experimental pump is arranged in a room.

FIG. 3 is a block diagram illustrating a flow of acoustic measurement and analysis.

FIG. 4 is a block diagram illustrating a method for obtaining a distance scale.

FIG. 5 is a graph showing an analysis result of a fan.

FIG. 6 is a graph of an analysis result of a pump.

[Explanation of symbols]

 Reference Signs List 10 fan 12 silencer 14 damper 20 pump 22 radiator 24 expansion tank 25 to 30 globe valve

Claims (3)

[Claims] 1. A method for detecting abnormalities of these devices by collecting sounds emitted from an air conditioning fan and a pump, comprising: a sound pressure signal in a normal state, a sound pressure signal in a normal state; In order to detect an abnormal signal by comparing with the above, after separating the signal into a low frequency range corresponding to the rotation frequency and a high frequency range corresponding to the natural frequency of the member, a filter by an AR model to which a linear prediction method is applied is used. An air-conditioning system using an acoustic method, wherein an abnormality of a fan and a pump is detected based on whether a value obtained by removing a characteristic of a sound pressure signal in a normal state from a sound pressure signal at the time of measurement exceeds a predetermined threshold value. An abnormality detection method for fans and pumps. 2. A prediction residuals when modeling normal signal Ss (n) of the AR model epsilon _ss (n) is, by the following equation prediction coefficients coefficients a _sk, a prediction order as p, Equation 1 ] Then apply the prediction coefficients a _sk abnormal signal Sf (n) determined the prediction residual epsilon _fs (n) of the following formula, Equation 2] Next, the ratio δ _fs of the predicted residual energies of the two is obtained by the following equation. Next, the ratio δ _sf of the predicted residual energy symmetrical to this is calculated by the following equation. 2. The method according to claim 1, wherein the sum of δ _fs and δ _sf is defined as a distance scale δa, and abnormality in the air conditioning fan and pump is detected by detecting a change in the distance scale δa. Wherein the modeled prediction residual epsilon _ss when normal signal Ss (n) of the AR model (n), calculated using the following expression prediction coefficients coefficients a _sk, a prediction order as p, Equation 5 ] Then apply the prediction coefficients a _sk abnormal signal Sf (n) determined the prediction residual epsilon _fs (n) of the following formula, [6] Next, the standard deviation of both prediction residuals is obtained, and the energy ratio δ ′ _fs of the prediction residual having an amplitude three times or more thereof is obtained by the following equation. Next, a symmetric ratio δ ' _sf of the predicted residual energy is calculated by the following equation. 2. The method according to claim 1, wherein the sum of δ ′ _fs and δ ′ _sf is defined as a distance scale δp, and the abnormality of the air conditioning fan and the pump is detected by detecting a change in the distance scale δp.