JP2000064964A - Failure prediction system for vacuum pump - Google Patents

Abstract

PROBLEM TO BE SOLVED: To predict in advance a clogging failure caused by a product precipitated in a casing of a mechanical booster type vacuum pump used in a water ring type vacuum pumping system of a semiconductor manufacturing apparatus, and replace the pump. To avoid product failure due to sudden pump failure, improve product yield,
Provided is a vacuum pump failure prediction system capable of reducing pump maintenance costs. SOLUTION: A sensor unit having an AE sensor 3 for detecting an AE (acoustic emission) generated by a vacuum pump 1, a current sensor 4 for detecting a current of a motor, and a temperature sensor 5 for detecting a temperature of an exhaust pipe. A monitoring system that collectively monitors a plurality of vacuum pumps 1 using signals from the sensor unit 2 and a diagnostic unit 15 that operates on a workstation 16 that connects the plurality of monitoring systems via a LAN (local area network).
And with.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a failure prediction system for a mechanical booster type vacuum pump, and more particularly to predicting a sudden failure of a mechanical booster type vacuum pump used in a water ring type vacuum exhaust system of a semiconductor manufacturing apparatus. The present invention relates to a vacuum pump failure prediction system that prompts replacement of a pump.

[0002]

2. Description of the Related Art A water ring type vacuum pumping system of a semiconductor manufacturing apparatus is
It is constituted by using a plurality of mechanical booster type vacuum pumps and a water ring type vacuum pump in the final stage for one semiconductor manufacturing apparatus (for example, a CVD apparatus). Explaining by taking a CVD apparatus as an example, among the gases used as materials for producing semiconductors inside the CVD apparatus, unreacted gases are evacuated by a vacuum pump of a water ring type exhaust system. Gas solidifies and precipitates in the exhaust pipe and inside the pump casing. When the deposited product accumulates, the rotor comes into contact with the casing, the pump is overloaded, and finally the pump stops. In the case of a batch processing type CVD apparatus, when the pump is suddenly stopped, the semiconductor being formed cannot be completed, which has a serious effect on the product yield. In practice, this is achieved by replacing the pump periodically according to the operating time. However, not only is the maintenance cost incurred, but catastrophic failures cannot be completely prevented by regular replacement alone.

On the other hand, an attempt has been made to monitor the operating state of the pump by measuring the motor current and the exhaust pipe temperature and detect a sign of a change in the operating state to predict a failure. It often rises immediately before, and even if an abnormal rise in current can be detected, sudden stoppage of the pump may not be avoided. Further, the exhaust pipe temperature is considered to increase due to the reaction heat of the gas flowing through the exhaust pipe, but it is difficult to find an accurate relationship between the amount of product deposited and the load state of the pump. Therefore, at present, clogging failure due to the product cannot be sufficiently predicted only by the motor current and the exhaust pipe temperature.

[0004]

SUMMARY OF THE INVENTION The present invention predicts in advance a clogging failure caused by a product deposited in a casing of a mechanical booster type vacuum pump used in a water ring type vacuum pumping system of a semiconductor manufacturing apparatus. Another object of the present invention is to provide a failure prediction system for a vacuum pump capable of avoiding a product failure due to a sudden failure of the pump, improving a product yield, and reducing a maintenance cost of the pump by prompting replacement of the pump.

[0005]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, the present invention is directed to a product generated by a product deposited inside a casing of a mechanical booster type vacuum pump used in a water ring type vacuum pumping system of a semiconductor manufacturing apparatus. A system for predicting a clogging failure, wherein an AE generated by a vacuum pump
An AE sensor for detecting (acoustic emission), a current sensor for detecting a current of the motor, and a temperature sensor for detecting a temperature of the exhaust pipe.
A monitoring system that collectively monitors a plurality of vacuum pumps using a signal from the sensor unit, and operates on a workstation that connects the plurality of monitoring systems via a LAN (local area network), and predicts a failure specific to the vacuum pump. And a diagnostic unit using a reference.

According to the present invention, a contact caused by a product deposited inside a casing of a vacuum pump in a water-sealed vacuum pumping system is detected by an AE, and a change in state of the vacuum pump is detected by a temperature and an electric current. By diagnosing the state of the vacuum pump using them and predicting a failure, a sudden failure of the vacuum pump is avoided in advance.

[0007]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A preferred embodiment of a system for predicting a failure of a mechanical booster type vacuum pump according to the present invention will be described below with reference to the drawings. The failure prediction system for a vacuum pump of the present invention predicts in advance a clogging failure due to a product deposited inside a casing of a mechanical booster type vacuum pump used in a water ring type vacuum pumping system of a semiconductor manufacturing apparatus such as a CVD apparatus. And a system for inviting pump replacement.

FIG. 1 is a block diagram showing the configuration of the apparatus of the present invention. As shown in FIG. 1, a failure prediction system for a mechanical booster type vacuum pump includes a sensor unit 2 that detects various states of a plurality (n) of vacuum pumps 1, and a measurement unit 6.
And a diagnostic unit 15.

(1) Sensor Unit The sensor unit 2 includes an AE sensor 3, a current sensor 4, and a temperature sensor 5 provided in n sets corresponding to the number n of vacuum pumps. AE (acoustic emission (acoust
ic emission)) The sensor detects contact between the rotor and the casing caused by products deposited inside the casing. The AE sensor 3 is installed on a part outside the pump casing via a sensor mounting stay. The current sensor 4 uses a current probe to detect a motor current from a power supply line of the pump driving motor. The temperature sensor 5 uses a thermocouple and detects the temperature of the pipe surface on the pump discharge side. Each of the sensors 3, 4, and 5 can be attached and detached without removing the pipe or stopping the operation of the pump. The sensors 3, 4, and 5 are collected as analog signals by the measuring unit 6 via a cable.

(2) Measuring Unit The measuring unit 6 holds an appropriate number of amplifier units 7, 9, 11 for amplifying signals from the n sets of AE sensors 3, current sensors 4, and temperature sensors 5, and holds signals. Data logger 1
2. It is composed of a personal computer 13 that controls each measuring instrument. The analog signal from the AE sensor 3 has an AE frequency band generated by the deposition of a product in the range of 250 kHz to 375 kHz.
0 is detected. Reference numeral 8 denotes an AE channel switch for switching a number of AE sensors 3. Each analog signal value of temperature and current is converted into a digital signal and held in the data logger 12, and the personal computer 1
Sent to 3.

The personal computer 13 collectively controls multi-channel sensors attached to a large number of vacuum pumps by controlling each measuring instrument, and periodically acquires data. At the time of data acquisition, the operation and stop of the pump are automatically determined from the pump current and the upstream and downstream pump current values. When the pump is restarted after the stop, the reference value of each measurement value is measured ( See below). The sensor unit 2 and the measuring unit 6 constitute a monitoring system. Although not shown in FIG. 1, a plurality of monitoring systems are installed in parallel.

(3) Diagnosis Unit The diagnosis unit 15 comprises a workstation 16 for connecting a plurality of monitoring systems via a LAN. The workstation 16 is a magneto-optical disk drive 1
7 and a warning light 18. The workstation 16 performs a diagnosis using the acquired data periodically sent from the measuring unit 6 of the monitoring system, and calculates a diagnosis output result of each pump, that is, a failure degree. A neural network is used for diagnosis. An embodiment of a method for determining each coefficient in a neural network will be described. First, 50 operation data from seven failures measured in advance are
Create zero input data and output data, that is, teacher data.

The input teacher data holds the maximum value from the start of operation after the pump replacement to the current time, the AE and the motor current represent a reference value, that is, the rate of change from the value immediately after the pump replacement, and the exhaust pipe temperature represents an absolute value. And normalize using the following equation: AEin = αAE · (AE1−AE0) / AE0 where AEin: AE effective value input teacher data AE1: Current value of AE effective value AE0: Value of AE effective value immediately after pump replacement αAE: AE effective value normalization coefficient = 0 ... Ampin = αAmp · (Amp1−Amp0) / Amp0 where Ampin: current input teacher data Amp1: current value of current Amp0: value immediately after pump replacement αAmp: current normalization coefficient = 0.5 Tmpin = αTmp · Tmp1 where Tmpin: temperature input teacher data Tmp1: current temperature αTmp: temperature normalization coefficient = 0.01

The above-mentioned normalization coefficient is appropriately determined according to the magnitude of the amount of change in data showing a significant tendency. The calculated input teacher data ranges from 0.0 to 0.1.
If it exceeds 0, it is set to 1.0. The failure degree, that is, the output teacher data is determined to be 0.6 or more at the time of failure. Next, the teacher data is trained by a back propagation method on a normal hierarchical neural network having three inputs and one output and an intermediate layer having one layer and one element, and a weight coefficient and a threshold value of the neural network are determined.

On the screen of the diagnostic system, a display similar to a display generally used by a failure prediction system of a plant using a window system (in which the arrangement of machines and a piping system can be listed) is displayed, and a water ring type vacuum exhaust system is displayed. Each pump is schematically displayed. The failure degree of each pump is indicated by five display colors. Clicking on the pump displays the operation history up to that point as a graph. Further, when a pump with a high degree of failure occurs or a system malfunction occurs, a warning light 18 separately provided is turned on together with a window display on the screen to urge the user to wake up.

FIG. 2 is a flow chart of measurement and diagnosis of the failure prediction system of the vacuum pump of the water ring type exhaust system shown in FIG. The personal computer 13 collectively controls the multi-channel sensors attached to the plurality of vacuum pumps 1 by controlling the measuring instruments to periodically acquire data (step 1). At the time of data acquisition, in step 2, the operation and stop of the pump are automatically determined from the pump and the upstream and downstream pump current values, and the pump is restarted after stopping (step 3).
, A reference value of each measurement value is measured (step 4). When the pump is operating, the display on the data collection unit is updated (step 5). After step 4 or step 5, detection of measurement abnormality or the like (step 6), creation of a diagnosis input value (step 7), diagnosis (step 8), and display update on the diagnosis unit side (step 9) are sequentially performed.

FIG. 3 is a diagram schematically showing the case where the reference value of each measurement value is measured. FIG. 3 (a) shows the AE effective value, FIG. 3 (b) shows the exhaust pipe temperature, and FIG. 3 (c) shows the pump current. 3A to 3C, the horizontal axis indicates time t, the solid line indicates the maximum value, and the broken line indicates the instantaneous value. Further, t1 indicates a pump stop time, t2 indicates a pump restart time, t3 indicates a reference value measurement time, and δt indicates a reference value calculation time. As is clear from FIGS. 3A to 3C, when the pump is restarted at a time t2 from the stop state, the pump is in a stable state after a lapse of a certain time immediately after the start. Therefore, the maximum value from the pump start time t2 to the time t3 after the elapse of δt is used as each reference value. That is, in FIG.
0 Is the AE reference value, and in FIG. Is a temperature reference value, and in FIG. Is a current reference value.

[0018]

As described above, according to the present invention,
Prevents clogging failures caused by products precipitated in the casing of a mechanical booster type vacuum pump used in a water ring type vacuum pumping system of semiconductor manufacturing equipment in advance, and prompts replacement of the pump, thereby causing sudden pump start-up. It is possible to avoid product defects due to failure, improve product yield, and reduce pump maintenance costs.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of a vacuum pump failure prediction system according to the present invention.

FIG. 2 is a flowchart of measurement and diagnosis in the failure prediction system of the vacuum pump according to the present invention.

FIG. 3 is a graph showing an outline of a reference value measuring method in the failure prediction system for a vacuum pump according to the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Vacuum pump 2 Sensor part 3 AE sensor 4 Current sensor 5 Temperature sensor 6 Measuring part 10 Effective value detector 12 Data logger 13 Personal computer 15 Diagnosis part 16 Workstation 18 Warning light

Continued on the front page (72) Inventor Manabu Mihokawa 4-2-1 Motofujisawa, Fujisawa-shi, Kanagawa, Japan Inside Ebara Research Institute, Ltd. (72) Inventor Fumomo Sato, 4 Kadotacho Industrial Park, Aizuwakamatsu-shi, Fukushima Prefecture Shares Fujitsu Tohoku Electronics Co., Ltd. (72) Inventor Hiroshi Takahashi 4th, Kadotacho Industrial Park, Aizuwakamatsu-shi, Fukushima Prefecture Inside Fujitsu Tohoku Electronics Co., Ltd. (72) Inventor Yoshiyasu Sato 4-1-1 Kamikodanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture F term in Fujitsu Limited (reference) 2G024 AD03 BA11 BA27 CA13 CA17 CA18 FA06 3H045 AA09 AA14 AA15 AA26 BA41 CA00 CA19 CA21 CA26 CA29 EA38 EA49

Claims (4)

[Claims] 1. A system for predicting a clogging failure caused by a product deposited inside a casing of a mechanical booster type vacuum pump used in a water ring type vacuum pumping system of a semiconductor manufacturing apparatus, wherein an AE generated by the vacuum pump is provided. (Acoustic Emission), a current sensor for detecting the current of the motor, and a temperature sensor for detecting the temperature of the exhaust pipe. A plurality of vacuums are provided by using signals from the sensor. A monitoring system that monitors the pumps collectively, and a diagnostic unit that operates on a workstation that connects a plurality of monitoring systems via a LAN (local area network) and uses a failure prediction standard specific to a vacuum pump. Features a failure prediction system for mechanical booster type vacuum pumps. 2. The sensor unit according to claim 1, wherein the sensor unit is detachable without stopping the operation of the vacuum pump during operation, which is a target of failure prediction, and without requiring any processing. Forecast system for mechanical booster vacuum pumps. 3. The values of AE, temperature, and current are measured within a certain period of time from the time when the vacuum pump starts operating, and those values are used as reference values to detect a rate of change from each reference value. 2. A failure prediction system for a mechanical booster vacuum pump according to claim 1, wherein failures are predicted by using the maximum values as diagnostic data. 4. A diagnostic unit that uses a failure prediction diagnostic criterion specific to the vacuum pump generates about 500 sets of input data and output data for teachers from operation data of the pump that has caused about 10 failures. 2. The weighting factor and threshold value of the neural network according to claim 1, characterized in that a three-input one-output intermediate layer is trained by a back propagation method on a one-layer one-element ordinary hierarchical neural network by a back propagation method. A failure prediction system for a mechanical booster type vacuum pump, wherein the teacher data is created by a specific method as described below, and the input teacher data is the maximum from the start of operation after the pump replacement to the current time. AE and motor current are the reference values, that is, the rate of change from the value immediately after pump replacement, and the exhaust pipe temperature is an absolute value. Normalized, AEin = αAE · (AE1-AE0) / AE0, where AEin: AE effective value input teacher data AE1: Current value of AE effective value AE0: Value immediately after pump replacement of AE effective value αAE: Specific to vacuum pump AE effective value normalization coefficient Ampin = αAmp · (Amp1−Amp0) / Amp0 where Ampin: current input teacher data Amp1: current value of current Amp0: value immediately after pump replacement αAmp: current normalization unique to vacuum pump Coefficient Tmpin = αTmp · Tmp1, where Tmpin: temperature input teacher data Tmp1: current value of temperature αTmp: temperature normalization coefficient peculiar to vacuum pump The normalization coefficient depends on the amount of change in data showing a significant tendency. The input teacher data calculated and determined appropriately is set in the range of 0.0 to 1.0, and when the input teacher data exceeds 1.0, it is set to 1.0. A failure prediction system for a mechanical booster type vacuum pump, which is determined to be 0.6 or more when a failure occurs.