KR100905235B1 - A precision diagnostic method for the failure protection and predictive maintenance of a vacuum pump and a precision diagnostic system therefor - Google Patents

Abstract

According to the present invention, the most challenging issue in the present invention has been to find a systematic method that allows maintenance engineers to determine the appropriate time for replacement of a vacuum pump based on their current performance test test results. In addition, the comparison of current diagnostic data with initial (or standard) data sets has been shown to allow a maintenance engineer to determine whether to replace a target vacuum pump based on derived pump performance indicators. In addition, the derived pumping speed indicators of current vacuum pumps are reported to be generally decreasing. These quantitative diagnostic analysis results are expected not only to determine the appropriate time for replacement of the vacuum pump based on their current performance performance results, but also to improve the stability and reliability of forecast maintenance of low vacuum pumps.

Description

A precision diagnostic method for the failure protection and predictive maintenance of a vacuum pump and a precision diagnostic system therefor}

The present invention relates to a precision diagnostic technique and a system for implementing the same for maintaining the forecast of a vacuum pump, and more particularly, to a semiconductor production process in various working conditions.

The demand for functionality and stability for vacuum pumps in modern semiconductor production processes has steadily increased. This demand is increased because as the size of the production wafer increases, the cost due to defective wafers and lost production time increases further. Technical requirements for vacuum pumps for such modern semiconductor processes are well pointed out by Bahnen and Kuhn [Ref. 1: R. Bahnen and M Kuhn, "Increased reliability of dry pumps due to process related adaptation and pre-failure. warning, "Vacuum, Vol. 44, No 5-7, pp. 709-712,1993]: high stability without pumping unplanned downtime, high water solubility to pump corrosive and reactive gas mixtures, high water solubility and low vibration, low noise to pump particulate and sublimable gas mixtures. In order to meet these demands, new dry pumps for modern semiconductor processes are not only suitable for various process-dependent operating conditions, but also the parameters associated with pump operation to prevent the risk of unplanned downtime (power, A monitoring plan should be provided for cooling water, purge gas and wear of pump parts-bearings, seals, gearboxes and motors. Bahnen and Kuhn [Ref. 1] proposed a plan to monitor process or operation-related parameters based on warning and alarm levels to prevent unexpected pump failures. However, it is process dependent and related to operation. All reasonable warnings for the parameters involved, and no reasonable measures for the level selection of the alarm have been presented. Such initial threshold selection is still a very challenging issue for early detection of failure of the vacuum pump. Moreover, the proposed monitoring plan does not provide any technical means for determining when to replace a vacuum pump that generates an alarm or warning signal. This is also a challenge for engineers performing vacuum pump repairs. Neither warning nor alarm are direct indicators of pump replacement, and only after investigating sufficient technical information on the performance and detected operating conditions of the determined vacuum pump can pump maintenance engineers decide whether to replace the pump. Because. The present invention not only provides a systematic way to approach the quantitative degradation of the vacuum pump that generated warnings and warning signals, but also enables pump maintenance engineers to determine whether to replace the pump based on the results of the performance assessment. Will be introduced.

Early level-based monitoring has been widely known as a traditional technique for protecting pumps from failures. [Reference 2: R.H. Greene and D.A. Casada, Detection of pump degradation, NUREG / CR-6089 / ORNL-6765, Oak Ridge National Laboratory, 1995]. However, Wegerich et al. [Reference 3: S.W. Wegerich, D. R. Bell and X. Xu, “Adaptive modeling of changed states in predictive condition monitoring,” WO 02/057856 A2, 2002; Reference 4: S.W. Wegerich, A. Wolosewicz and RM Pipke, “Diagnostic systems and methods for predictive condition monitoring,” WO 02/086726 A1, 2002, pointed out the shortcomings of an initial warning and alarm plan based on the output of the sensor: " Existing technologies have not responded to global changes in the operating parameters of a process or machine, and sometimes failed to provide adequate warnings to prevent unexpected shutdowns, equipment damage or significant safety risks. " In order to overcome these limitations in the existing technology, they are based on neural networks suitable for new operating conditions [Ref. 3] and analysis systems based on these models for monitoring predictive conditions. It proposed the use of [Ref.4]. As is known from previous studies on emotion and coordination of dynamic systems [Reference 5: Wan-Sup Cheung, “Identification, stabilization and control of nonlinear systems using the neural network-based parametric nonlinear modeling,” Ph.D. Thesis, University of Southampton, 1993], neural network models are useful for interpolating new states that lie between training data sets and for extrapolating adjacent states that are very close to the trained set. Have the ability. Wegerich et al. [Ref. 3, Ref. 4] utilize the learned neural network insertion and extrapolation capabilities [Ref. 5] to estimate the current state of a process or machine corresponding to the sensor's output measurements. In addition, the error between the estimated state value and the measured state value can cause an early warning of an error, perform statistical tests to investigate changes when the process or system is operating under new conditions, or for new operating areas. It is used for the reconstruction of the learning set. The proposed signal processing plan, which includes the design of a new learning set for a modified operating area and a learning set for a model learning process of the learning set, and which generates an alarm and is suitable for the modified operating area, requires not only difficult computational tasks, Even the inherent complexity of the proposed model-based analysis system appears to be involved. This complexity of computational output and execution of the proposed inspection system has become an inevitable technical issue encountered in pump observation and analysis systems for modern semiconductor fabrication processes. Moreover, the proposed model-based analysis system does not provide any systemic way of evaluating the performance of vacuum pumps operating under various operating conditions. As a result, these technical issues have become the main motivation of the present invention to develop not only a simple model suitable for pump operating conditions, but also a new evaluation plan of vacuum pump performance indicators applicable to the place where the pump is installed. The present invention proposes a forecast maintenance plan for a vacuum pump that can estimate pump performance indicators whenever a warning or alarm signal is observed. This direct performance evaluation plan does not require any of the learning sets and learning models proposed by Wegerich et al. [Ref3, Ref4].

Instead of using the above parameter model suitable for changing the operating conditions of the vacuum pump over time, Ushiku et al. [Ref.6: Y.Ushiku, T.Arikado, S.Samata, T.Nako, and Y.Mikata, "Apparatus for predicting life of rotary machine, equipment using the same, method for predicting life and determining repair timing of the same," US Patent Application Publication, US2003 / 0009311 A1,2003], Samata et al. [Ref.7: S.Samata , Y.Ushiku, K.Ishii, and T.Nakao, "Method for diagnosing life of manufacturing equipment using rotary machine," US Patent Application Pulblication, US2003 / 01543997 A1,2003] and Ishii et al. [Ref.9: K. Ishii, T. Nakao, Y. Ushiku, and S. Samata, “Method for avoding irregular shutoff of production equipment and system for irregular shutoff , "US Patent Application Publication, US2003 / 0158705 A1,2003] A statistical analysis method and mahalanobis distance for determining whether current measured time series data is different from a standard data set of time series values corresponding to normal operating conditions. Based on the proposed analysis method [Ref.10: WH Woodall, R, Koudelik, Z.G.Stoumbos, K.L. Tsui, S.B. Kim, CP. Carvounis, "A review and analysis of the Mahalanobis-Taguchi system," TECHNOMETRICS, Vol. 45, No. 1, pp 1-14,2003]. The statistical methods are based on secondary statistical characteristics of sampled signals such as mean value, standard deviation, correlation, etc. (Ref.11: JSBendat AGPiersol, Random data: Analysis and measurement procedures, John Wiley & Sons: NY, 1985). It is based. Since statistical properties can only be used in normal processes, such a scheme has limited applicability to the various output dependent operating conditions required for other products. This means that for each output dependent task a standard data set of each time series value is required. A major issue here is how to design a standard data set of output dependent time series values sufficient to cover all ranges of normal operating conditions. No efficient way of designing such data sets has been proposed by Ushiku et al. [Ref. 6], Samata et al. [Ref. 7, Ref. 8] and Ishii et al. [Ref. 9]. Although time series data at normal operating conditions for new or repaired vacuum pumps are available only near the beginning of each planned process, it is standard for the full range of normal operating conditions without time-consuming data collection and signal processing operations. Data could not be obtained. Indeed, the manufacturing units of modern semiconductors require a variety of processes at individual operating conditions such as various camber pressures, gas flow rates and gas mixtures, and gas characteristic values. The process-related characteristics and operating conditions in these semiconductor fabrications are very confidential to the supplier of the vacuum pump. It is important that the vacuum pump observation and analysis system for modern semiconductor processes be self-adapting to various operating conditions. Since the proposed statistical analysis [Ref 6-9] does not take into account any systemic basis for obtaining performance indicators for vacuum pumps, this approach allows the engineer to determine the appropriate timing for pump replacement. Note that it does not present any quantitative data on. The present invention has been proposed to provide a way to realistically solve such technical problems without using standard data widely collected under normal operating conditions utilized in the prior art [Ref 6-9].

The inventors of this technique have already developed an accurate performance test and evaluation method for low vacuum pumps and have published their experimental results in several technical papers [Ref. 12: JYLim, SHJung, WSCheung, KHChung, YHShin]. , SSHong, and WGSim, "Expanded characteristics evaluation for low vacuum dry pumps," AVS 49th International Symposium, xx, 2002; Reference 13: JYLim, WSCheung, JHJoo, YOKim, WGSim, and KHChung, "Characteristics evaluation practice of predictable performance monitoring for low vacuum dry pumps," AVS 50th International Symposium, 9-10, 2003; Reference 14: W. S. Cheung, J. Y. Lim and K. H. Chung, "Experimental study on noise characteristics of dry pumps," Inter-noise 2002, Port Lauddale: USA, 2002; Reference 15: W. S. Cheung, "Acoustic characteristics of dry pumps designed for semiconductor processes," Inter-noise 2003, Jeju, Korea, 2003]. Such experiments were performed by a low vacuum pump test bench, the overview of which is shown in FIG. 1.

The test bench is used to determine the performance factors of low vacuum pumps such as pumping speed (volume flow rate), ultimate internal pressure, compression ratio, gas output simulation, maximum and minimum values of operating pressure, power consumption, residual gas analysis and mechanical noise, vibration levels, etc. Has been. Hundreds of vacuum pumps supplied to semiconductor manufacturers have been tested. The test results of these vacuum pumps provided a systematic understanding of the key performance factors and dynamic characteristics of the vacuum pump.

In order to obtain pumping speeds, the present invention can be most widely utilized in vacuum pumps and utilizes with complete accuracy a perfect manner that can be applied to actual operating pressure ranges and pump capabilities.

Square, asterisk, and circular notation lines represent the maximum, minimum, and average values of the test results of the pumping speed, respectively, which test individual gas simulated by adjusting the internal gas pressure of the test dome shown in FIG. Obtained from output conditions. The coefficient of variability of the pumping speed, defined as the ratio of the standard deviation to the mean value, is found to be 6.7% when the internal pressure of the test dome is equivalent to 0.01 mbar, and 6.7% corresponding to an internal pressure of 0.002 mbar. It became.

At levels above 0.05 mbar, the coefficient of variation was found to be 3.5% or less. This means that within a small variation, the quality control of the pumping speed of the tested vacuum pump is very good. It is noteworthy that the pumping speed with small variability is a good indicator of how much the current performance of a vacuum pump has been reduced. Indeed, the pumping speed is the most important factor among the performance parameters of low vacuum pumps. The present invention proposes the use of pumping speed as a new state variable. However, previous inventions for observing the operating conditions of vacuum pumps did not consider the pumping speed as the observed state variable. In the next section, the present invention even proposes a systemic method for obtaining pumping speed indicators in a semiconductor fabrication factory where a pump is already installed, which is an in-situ estimation method. The estimated pumping speed indicator allows the pump maintenance engineer to determine the appropriate time to replace the pump, so the estimated pumping speed indicator has been shown to play an important role for accurate forecast maintenance of the vacuum pump.

3 shows how much mechanical noise and vibration level variation is present between the tested pumps. Although the pumping speeds of the tested pumps had small variability as shown in FIG. 2, the mechanical noise and vibration levels were very different for each pump of the same model. The mechanical noise level is estimated by taking the average of the sound pressures measured from the ten positions recommended by the ISO 3744 standard. The maximum difference in the mechanical noise level values was observed to be 12 dBA when the test dome pressure was 2 mbar. Under other gas output conditions, the difference in sound pressure level value SPL approached about 9 dBA. This large SPL difference corresponds to a 4 times loudness difference (2 times the loudness per SPL difference of 5 dBA). The coefficient of variation of the sound pressure level was found to be 51% to 65% when the pressure range was 0.01 mbar to 10 mbar or more. The coefficient of variation of the mechanical vibration (acceleration) level was found to be 19% to 23% when the pressure was below 1 mbar and increased to 51% as the gas pressure reached 10 mbar. Moreover, the ratio of the minimum level value and the maximum level value of the mechanical acceleration level was 1.4 to 1.6 when the gas pressure was 1 mbar or less, but it was observed that the ratio increased rapidly to 3.3 as the gas pressure reached 10 mbar. This great change points out that each individual pump has its own normal operating conditions for mechanical noise and vibration. This pump-by-pump working characteristic can lead to unstable and inconsistent state observations, such as false alarms and alarm signals, even for a normally functioning instrument. The use of a fixed level based initial level threshold for generating a warning or alarm signal for the system has caused great difficulty.

In order to improve such limited capability of fixed level based machine condition observation and diagnostic systems, the present invention will propose an active algorithm which is self-adaptable to normal operating conditions dependent on pumps in the next section. Detected faults in vacuum pumps such as wear of mechanical parts, such as ball bearings, journal bearings, gears, pump lobes, seals, sizing of rotating elements, etc. Mechanical vibration and noise signals are also utilized in the present invention for the implementation of. The above mechanical defect is well diagnosed from the spectral analysis of the mechanical vibration or noise signal proposed in the previous study [Ref.2]. Of course, mechanical defects also make it possible for pump maintenance engineers to determine the proper time for pump replacement.

It should be noted that the gas output dependent state variables of this machine operated observation system are not limited to noise and mechanical vibration signals. 4 shows the statistical characteristic values (maximum value, minimum value, average value) of the power consumption values measured from the booster pump and the dry pump. The maximum and minimum ratio of power consumption for the booster pump was 1.3 at a gas pressure of 2 mbar or lower, and increased to 1.6 as the gas pressure reached 10mba. The coefficient of variation of the booster pump was 9% to 11% when the gas pressure was below 1 mbar, but rose steeply to 57% as the gas pressure reached 10 mbar. In contrast to this large deformation of the power consumption of the booster pump, the maximum and minimum ratio of the power consumption value of the dry pump was observed to be 1.1 to 1.2 within the pressure range of the test. The coefficient of variation was also found to be 4% to 6% within the test pressure range. These test results indicate that the total power consumption value is useless for the state observation system because the total power consumption value of the booster pump and the dry pump is a highly variable state variable. As a result, two separate power consumption values of the booster pump and the dry pump will be considered in the present invention.

It is very important to understand how much the measured state variable values increase as the gas output conditions vary in the pump operating area. The experimental results presented in FIGS. 3 and 4 help to find a solution to this task by carefully observing the mean values (indicated by the solid lines marked with asterisks). Although the gas pressure in the test dome gradually increases to some extent, the average value remains the same. This is the area where the measured state values of mechanical noise and vibration and power consumption levels are independent of the gas output. The present invention does not utilize characteristics independent of gas output among the state variables measured for analyzing the operating conditions of the vacuum pump. Such gas output independent conditions are very often found in actual process conditions. A good example is the "idle" state of a working vacuum pump, in which no external gas is supplied to the inlet of the pump. In the next section, the present invention will propose a systemic method for modeling the output independent characteristics of the state variables of the vacuum pump observation and diagnostic system.

Moreover, as the gas pressure increases in a region independent of gas output, the mean values of mechanical noise, vibration and power consumption parameters increase. For example, the maximum value of the mechanical noise level in the gas output dependent region appears to be 12 dBA (4 times) higher than the maximum mechanical noise level in the region independent of the gas output. Similarly, the maximum value of the mechanical vibration level is 2.4 times higher in the region depending on the gas output, and the power consumption level values of the booster pump and the dry pump are 2.3 times and 1.2 times higher, respectively. Here, another technical issue encountered in state observation and diagnostic systems is finding a suitable model to account for such gas output dependent characteristics of state variables, which means that the actual working area of the vacuum pump always includes a gas output dependent condition. Because. In the next section, the present invention will also propose a systemic way of modeling the dynamic properties of state variables in the region depending on gas output. Of course, mathematically identical models have been shown to be applicable to both gas independent and dependent conditions. As a result, one model will be provided for the operating area independent of the gas output and the other model will be provided for the area dependent on the gas output. The use of two separate models has been developed to increase the stability and reliability of detecting abnormal operating conditions as soon as possible in a vacuum pump.

In the present invention, the observed information regarding the gas output condition, such as the inlet gas pressure signal of the vacuum pump, is characterized by distinguishing abnormal operating conditions of the vacuum pump, and more specifically, the increase in the observed state variables due to the output of the gas. It is very clear that it plays an important role in judging whether or not it is. In order to improve the ability to more reliably diagnose abnormal operating conditions of the vacuum pump, using the pressure information of the intake gas has not been devised in the previous inventions. In the present invention, observing the pressure of the intake gas seems to enable the quantitative analysis of the pumping speed with the improvement of the diagnostic ability. The above point is very important because finding an indication of the pumping speed leads to a suitable time to replace the target vacuum pump with a new one. The present invention provides a reasonable way to find the pumping speed of a vacuum pump operating in a semiconductor fabrication plant. Estimated pumping speed indicators enable pump maintenance engineers to determine the appropriate time for pump replacement, and these indicators appear to play an important role in maintaining accurate forecasts for vacuum pumps.

Technical solution

According to the present invention, there is provided a precise analysis method for the failure prevention and forecast maintenance of the vacuum pump, including the following steps.

1) collecting state variables associated with various pump operations of the newly installed vacuum pump at preset sampling rates for idle conditions and various gas output conditions.

2) the maximum of the time series values of the state variables from each set of consecutively sampled signals over a period of time adopted by the user, which is longer than the dominant period of the state variable signal element which fluctuates under idle conditions and various gas output conditions. And determining the minimum value

3) estimating pump operating characteristic values using an active diagnostic algorithm based on parametric models under idle conditions and various gas output conditions.

4) obtaining the pump performance index of the newly installed vacuum pump by using an in-situ evaluation method

5) storing the estimated pump operating characteristic value and the pump performance evaluation index of the newly installed vacuum pump in the vacuum pump maintenance database

6) repeating steps 1 to 5 whenever the newly installed vacuum pump is observed to be in abnormal operating conditions

7) The estimated pump operating characteristic value and the pump performance evaluation index of the newly installed vacuum pump stored in the vacuum pump maintenance database to determine whether the pump is replaced, the estimated pump operating characteristic value and the pump performance under abnormal operating conditions. Steps to Compare with Evaluation Indicators

According to the present invention, there is also provided a precise diagnosis system for failure prevention and forecast maintenance of the vacuum pump consisting of the following components.

A dedicated signal conditioning unit for amplifying suction pressure gauges (or vacuum gauges), exhaust pressure transducers, booster pumps and supply pumps of dry pumps and mechanical vibration sensors and measurement microphones centrally located in booster pumps and dry pumps;

A high speed multichannel data acquisition (DAQ) system suitable for collecting mechanical vibration and sound pressure signals with ultrahigh frequency components ranging from 10 kHz to 20 kHz;

A dual processor server-class PC system with sufficient performance to allow for steps 1 to 5 below;

1) Acquiring all state variable values measured from a real-time DAQ system without losing any data transmission. 2) Obtaining the root mean square of mechanical vibration and sound pressure signals. 3) Idle conditions and other gas output conditions. Determining the maximum and minimum time series values of the measured state variable values from each group of successively sampled signals in U. 4) Using a model-based active diagnostic algorithm, optimized model parameters and the measured state Estimating pump automatic characteristic values consisting of average and maximum values of the parameters. 5) Obtaining pump performance indicators of the newly installed vacuum pump using an in-situ evaluation method, and estimating the estimated performance of the newly installed vacuum pump. Pump operating characteristic values and pump performance evaluation indicators are stored in the vacuum pump maintenance database, The estimated pump operating characteristic value and pump performance evaluation index of the newly installed vacuum pump stored in the vacuum pump maintenance database to determine whether to replace the pump, and the estimated pump operating characteristic value and pump performance evaluation index under abnormal operating conditions and Comparison step

Favorable effect

The most challenging issue in the present invention has been to find a systemic way that allows a maintenance engineer to determine the appropriate time for replacement of a vacuum pump based on the results of the current performance evaluation of the vacuum pump.

As described in the 'Technical Problems' section, the performance test results for a large number of low vacuum pumps convinced the team that they could not be solved without addressing two fundamental issues. The first issue is to improve the low level of stability and consistency of the diagnostic results with measured state variable values, which are mainly caused by individual pump by pump operating characteristics and various process conditions. The second issue is the realization of outdoor (or field) performance testing of low-vacuum pumps in places where pumps are installed, rather than well-equipped test labs.

To approach the first issue, an active algorithm based on a linear parametric model is proposed. The linear parametric model is believed to represent the upper and lower asymptotic boundaries of the dynamic (varying amplitude) characteristics inherent in the measured state variables. The proposed active algorithm provides model parameters optimized for the operating conditions (ie multiple process conditions) of individual vacuum pumps as well as pump by pump dependent operating characteristics (i.e. large variety of vacuum pump operating characteristics). Seems to do. Detailed digital signal processing schemes have proved very successful in estimating the operating characteristic values of a vacuum pump from measured signals of six state variables placed in a semiconductor process.

To address the second issue, the present invention proposes a simplified version of a pump down test method developed for evaluating vacuum pump performance as a precision diagnostic analysis required for forecast maintenance of low vacuum pumps. do. The outdoor performance evaluation plan, also referred to in the proposed field performance evaluation scheme, consists of obtaining four performance indicators of the pump for a virtual gas output field test under conditions close to the actual process and a test condition dependent on each gas output. In the same way to obtain the above operating characteristic values, the pump performance indicators for each of the gas output operating conditions and the idle conditions are obtained. It is evident that the present invention has developed the results of combining the pump performance indicators with the work process characteristic values of each measured state variable in order to perform a precise diagnostic analysis for the maintenance of a vacuum pump. The combined results obtained from the newly installed vacuum pump are shown in detail in the Optimal Mode section, which is used as the initial (or standard) data set of the tested vacuum pump for further diagnostic analysis to be performed in the future. .

Furthermore, by applying the on-site diagnosis plan of the vacuum pump performance to a vacuum pump that has reached an abnormal operating condition, that is, a condition that responds slowly to the intended vacuum level of the reaction chamber, it is developed for precise analysis for maintaining the forecast of the vacuum pump. The feasibility and effectiveness of the on-site diagnostic plan of the vacuum pump performance is investigated. Operating characteristic values and pump performance indicators derived from abnormal operating conditions are detailed in the Optimum Mode section. The comparison of the current diagnostic analysis results with the original (or standard) data set seems to enable the maintenance engineer to determine whether to replace the target vacuum pump based on the obtained pump performance indicators. In more detail, the obtained pumping speed indicator value of the current vacuum pump is reported to decrease by 31%. These quantitative diagnostic analysis results not only allow the maintenance engineer to determine the appropriate replacement time for the vacuum pump based on the current performance results of the vacuum pump, but also improve the stability and reliability of low vacuum pump forecast maintenance. Let's do it.

It should be noted that in the present invention, the inlet pressure signal measured as a standard state variable plays a central role in obtaining the proposed pump performance indicators. This not only enables the estimation of pumping speed indicators, but also appears to lead to quantitative analysis of gas output dependent operating characteristics defined by three other performance indicators related to the exhaust pressure and the current supplied to the booster and dry pump. Moreover, the inlet pressure signal is also utilized as a standard state variable that classifies the current pump operating state as a gas output state or idle state. The two states separate operating characteristic values, i.e., values estimated under idle operating conditions of the vacuum pump and those estimated under gas output operating conditions. This means that the measured signals of the state variables under two operating conditions have very different statistical characteristics, and separate trend observation and diagnostic analysis for idle and gas output operating conditions realize the performance improvement for early detection of failure of the vacuum pump. For it was chosen. The present invention also proposes a rational way of designing a series of operating characteristic values derived from each measured state variable into matrix-type data suitable for multivariate statistical analysis, performance analysis, and Mahalanobis distance analysis. . Porting the data matrix of the model parameter structure into such classical analysis algorithms (multivariate statistical analysis, process performance analysis and Maharanobis distance analysis) is of course one of the major achievements provided by the present invention. The proposed active diagnostic algorithm was developed to provide predictive maintenance as well as to realize early detection of a degraded vacuum pump to protect the pump from failure.

Finally, the present invention proposes a realistic implementation system for precise diagnostic analysis for predictive maintenance of low vacuum pumps. This system appears to include six types of state variable measurement sensors, signal conditioning amplifiers corresponding to the sensors, 16 channels of high speed data acquisition systems and server class PC systems. As shown in the Optimal Mode section, the developed transition system is successful in precise diagnostic analysis for forecasting low vacuum pumps, so a compact version of the transition system that is portable and mobile at the actual process site can be supported. .

1 shows a schematic diagram of a performance test bench of a low vacuum pump according to the present invention.

Figure 2 shows the statistical characteristics for the pumping speed of the low vacuum pump.

3A and 3B show the noise level characteristics and the mechanical vibration level characteristics of the spatially averaged low vacuum pump, respectively.

4A and 4B show power consumption characteristics of the booster pump and the dry pump, respectively.

5A-5D show measured state variable signals, suction pressure and exhaust pressure, and supply current, respectively, for boosters and dry pumps.

6A-6D show a comparison of the classified maximum, minimum (thin solid line) of the inlet and exhaust pressures of the booster and dry pump and current signals and the estimated results (solid solid line) based on a suitable model, respectively.

7A to 7D show the rms values of the vibration acceleration and the noise signal and their upper and lower asymptotic boundary curves (thick solid lines), respectively.

8A and 8B show the suction pressure signals of the first and second sub-progression zones existing between the first gas output operating region and the second idle region. (Bold solid lines represent a suitable model of the exponential decay function.) .)

9A-9F show state variable signals measured from vacuum pumps under abnormal operating conditions.

FIG. 10 shows the suction pressure in the sub-progression region between the gas output operating region and the idle region shown in FIG. 9A.

11 shows a schematic layout of the state variables measured by a system of implementation of a precision diagnostic analysis for the maintenance of low vacuum pumps.

Fig. 12 shows a system for performing a precise diagnostic analysis for forecast maintenance of a low vacuum pump installed at a semiconductor fabrication site.

The most appropriate mode for the implementation of the invention

Precise method for maintaining forecast of low vacuum pump

Whenever a warning or alarm signal is generated by a low vacuum pump, it is very often seen in semiconductor manufacturing plants that most maintenance engineers have a great difficulty in determining the appropriate time to replace a low vacuum pump. This is because the early warning or warning signal does not always indicate when the replacement of the pump that caused the warning or warning signal has expired. Without technical information on the pump's detailed performance results, no maintenance engineer can replace the pump that generated the signal with a new one. However, conventional observation or diagnostic systems do not provide a systematic approach to obtaining performance indicators of vacuum pumps operating at alarm or warning level conditions. Although the modified version of the pump performance test, as presented in the previous section, does not fully meet all the elements of the international standard for the performance testing of low vacuum pumps, it is not forbidden even where the pump is installed. It was obvious. The present invention proposes an on-site means for obtaining a performance index of a vacuum pump installed in a semiconductor factory.

The present invention considers the normal (or initial) operating conditions to be the operating state of the new or reconditioned vacuum pump which has completed the warm up phase after factory installation. This indicator is very useful because the performance indicators obtained from the initial conditions enable the maintenance engineer to determine how different the operating conditions are from the initial state, eg at the moment when an alarm or warning signal is observed. It is obvious. When a new or reconditioned pump is ready for normal operation, the present invention provides a pump for individual gas output conditions such as 25%, 50%, 75%, 100% levels of the maximum pressure of the reaction chamber expected in the actual process. It is suggested to perform a performance test. When chamber pressure levels subdivided from process conditions are possible, such pressures are also selected for the gas output conditions of the gas output dependent performance test. The characteristics of the state variables for the initial operating conditions are also utilized for trend observation and diagnostic analysis of the target pump, so the same gas output test for the process conditions is expected to be the most appropriate choice. Since nitrogen gas can be used for the nominal test of low vacuum pumps, it is also recommended to use nitrogen gas instead of reprocessing gas. Note that the initial gas output performance test is also used to obtain the gas output dependent performance of the newly installed vacuum pump as well as to obtain the dynamic characteristics of the initial state variables. The invention also uses the characteristics of the initial state variable measured from the gas output operating conditions to determine how different the operating conditions are from the initial state, such as at the moment when an alarm or warning signal is observed. In order to realize the precise diagnostic analysis for the maintenance of the low vacuum pump, both the dynamic characteristics of the state variable and the obtained performance index are considered in the present invention.

In order to accurately describe the dynamic characteristics inherent in the measured state variables, digital signal processing and adjustment theory [Ref. 16: B.Windrow and S.D. Steams, Adaptive Signal Processing, Prantice-Hall, Englewood Cliffs: NJ, 1985; Ref. 17: P.A. Nelson and S.J.Elliott, Active Control of Sound, Academic Press, London, England, 1992] select an active algorithm based on a parametric model. Active algorithms enable dynamic estimation of model parameters that are well suited to various state variables. The estimated model parameters are utilized to diagnose the operating conditions of the vacuum pump, ie to quantitatively determine how different the operating conditions are from the initial conditions. The above theoretical approach is referred to herein as an "active diagnostic" algorithm. Active diagnostic algorithms are the subject matter of international patents pending in the name of the same inventors, which are also developed in the present invention. It is important to note that the active algorithm provides a set of model parameters tailored to individual pump operating conditions, such as multiple process conditions. Of course, active algorithms still enable estimation of model parameters for various pumps. This set of pump dependent parameters is extremely useful for investigating the operational variations of the same model vacuum pump group. This is why an active algorithm based on a parametric model is used to diagnose vacuum pumps.

1. Active Algorithm for Parameter Model of Diagnostic State Variables

The state variable according to the invention is defined as one of the periodically sampled physical properties chosen for quantitatively examining the operating conditions of the target vacuum pump. In relation to pump operation, the suction and exhaust pressures, the motor supply current to boosters and dry pumps, mechanical vibration signals, negative pressure signals, purge gas pressure and gas flow rate, body temperature, coolant temperature, lubricant pressure, dry pump There are various state variables such as level. The first step in maintaining the forecast of the vacuum pump is to collect an initial data set representing the mechanical properties of the above state variables for the individual gas output conditions. To collect them, a newly installed vacuum pump was selected that was ready for normal operation. The process gas output level required for the vacuum pump is known to correspond to two suction pressures of 10 mbar and 14 mbar, respectively. Performance test results obtained from the other two gas output conditions are shown in this section.

5 (a) to 5 (d) show the measured state variable signals, where 5 (a) is the suction pressure, 5 (b) is the exhaust pressure, and 5 (c) is the supply current to the booster pump, 5 (d) shows the supply current to the dry pump, which was sampled at 10 words per second. As shown in Fig. 5 (a), two different suction pressure levels match well with the intended levels. The two high amplified level regions correspond to the pumping state. The base pressure level of the inlet corresponds to the idle operating state of the vacuum pump, in which any pumping gas is supplied to the external minimum to the process chamber. It is evident from Figs. 5 (b) to 5 (d) that the exhaust pressure and the current supplied to the booster pump and the dry pump depend on the suction pressure level such as the gas output condition. These values are good examples of gas output dependent state variables. The basic question here is how to explain the observed mechanical properties from gas output dependent state variables. Observation of the mechanical properties shown in FIG. 5 leads to the selection of a parametric model in the present invention. In order to explain the range of the variable amplitude signal, the upper and lower asymptotes are considered in the present invention. The method of modeling the amplitude range has already proved to be very efficient for trend observation and diagnostic analysis of vacuum pumps, as demonstrated in the previous invention [18]. To clarify the subject matter of the present invention, an overview of the implementation of the amplitude range modeling approach has been changed in this section.

Let ym denote the m-th sampled suction pressure signal and the subscript m denote the time index. In the present invention, the sampling rate was selected to 10 Hz (10 samples per second). As shown in FIG. 5, sampled time series values {ym: m = 1,2, ..} of each measured state variable are used to select the maximum and minimum values on a user-selected period, ie every 20 seconds. The period selected by the user may be selected to be longer than a period of a slowly varying signal, that is, a period of a dry pump (DP) supply current shown in FIG. 5 (d).

By using FFT (Fast Fourier Transform) analysis, the dominant period of the varying DP supply current signal shown in FIG. 5 (d) was found to be close to 20 seconds. As a result, classification of the maximum and minimum values respectively was performed on the recorded signal for each 20 seconds. The second half (50%) of the signal recorded for 20 seconds overlapped the classification of each set of maximum and minimum values obtained for every 10 seconds. In the present invention, in order to examine how the average value differs from the classified maximum and minimum values, the average value of the signal recorded for 20 seconds is additionally obtained. In addition, the peak values during each idle state and gas output state are also investigated. The peak values corresponding to the repeated idle and gas output conditions are utilized to investigate how many unexpected changes in each state variable occur in each idle condition and gas output condition.

6 (a) to 6 (d), 6 (a) is an intake pressure signal, 6 (b) is an exhaust pressure signal, and 6 (c) is a supply current signal to the booster pump. (d) shows the comparison of the maximum and minimum values (thin solid line) categorized for each variable in the supply current signal to the dry pump and the estimation results based on the appropriate model (bold solid line). The solid circle on the thin line indicates the peak value at idle and gas output operating conditions.

The classified maximum and minimum values are shown in Figures 6 (a) to 6 (d). The maximum and minimum values of the suction pressure signal are taken as {yU, n, yL, n: n = 1,2, ...} from each set of consecutively sampled signals in the first gas output operating interval. The present invention proposes a linear model for explaining the upper and lower asymptotes given below.

In Equation 1, the subscript k represents an upper or lower asymptote model, i.e. k = U for the upper model and k = L for the lower model. In Equation 1, two sets of {a _k , b _k : k = U or L} model parameters are easily obtained by using the least squares method. In the first gas output state let the time series values of the classified maximum and minimum values be {y _{k , n} = 1,2, ... N}. The optimized model parameters are found as follows.

The first parameter {a _k : k = U or L} in Equation 2 is the slope of the suction pressure signal indicating the increase or decrease rate. The second parameter {b _k : k = U or L} indicates each initial suction pressure level (ie the value at n = 0). The thick solid line in FIG. 6 shows the values obtained from a suitable model for the upper and lower asymptotes. The estimated upper and lower asymptotes appear to be appropriately formed at the upper and lower boundaries of the measured amplitude signal. Moreover, the estimated model parameters can be used to investigate how much change in suction pressure is present at the first gas output operating condition. This indicates that the trend of suction pressure can be quantitatively characterized by the estimated model parameters. The above points are of great value, as two sets of appropriate model parameters enable trend observation and diagnosis for the measured suction pressure conditions. Since the proposed approach does not use all the set of sampled time series data, the use of appropriate model parameters provided a lot of memory savings. This means that implementation systems based on small hardware can be realized by using appropriate model parameters. It should be noted that the mean and standard deviation at each of the upper and lower asymptotes can be obtained by using the equation

If the slope value is zero (a = 0 in Equation 3), the second parameter appears to be an average value. As shown in Equation 3, the estimated parameters do not only enable the calculation of statistical characteristic values (average value and standard deviation value) for the estimated model. This indicates the usefulness and efficiency of using a parametric model that is suitable for the statistical properties of the measured state variables. As shown in Fig. 5 (a), the magnitude of the suction pressure appears to be uniform, but there appears to be some variation when the magnitude of the even amplitude level zone is reduced. As shown in Fig. 6 (a), the estimated model parameters appear to be sufficient to investigate how well the suction pressure is maintained under gas output conditions.

Given the categorized maximum and minimum time series values of the sampled suction pressure signal in the first idle state, two sets of upper and lower model parameters are obtained using equation (2). The mean value and standard deviation of each asymptote are also obtained from Equation 3. The estimated set of parameters also appears to provide enough information to determine how much vacuum level can be maintained at the inlet of the tested pump. Similarly, model parameters of the upper and lower asymptotes are estimated at different idle conditions and gas output conditions. Suitable model parameters and the statistical properties of these parameters under idle conditions and gas output conditions are used for forecast maintenance of the vacuum pump. The combined set of parameters in the idle and gas output operating conditions not only examines how many gas output conditions have been applied on the vacuum pump, but also helps determine how much vacuum level is maintained in the idle state. useful. Knowledge of the gas output conditions of the vacuum pump will appear to play a very important role in determining when to replace the vacuum pump properly under abnormal operating conditions. The present invention places high emphasis on the use of suction pressure signals for forecast maintenance of vacuum pumps that are used in detail in semiconductor fabrication processes.

The theoretical background of the parametric model, adopted to explain the mechanical properties of the suction pressure signal observed from the vacuum pump, has been mentioned so far. This approach can also be applied to other state variables, such as the exhaust pressure and supply current signals of booster pumps and dry pumps shown in FIGS. 5 (b) to 5 (d). Time series of maximum and minimum values of each state variable value The value can be easily obtained by classifying the maximum and minimum values from a block of 200 consecutive samples each time supplied from the data acquisition system (which corresponds to a signal recorded for 20 seconds). 6 (b) to 6 (d) show the estimated results based on the time series values (thin solid line) of the classified maximum and minimum values and the appropriate models for the exhaust pressure signal and the current signals supplied to the booster pump and the dry pump, respectively. Thick solid line). Given the classified time series values of each state variable in the idle and gas output operating conditions, two sets of parameters corresponding to the upper and lower boundaries are obtained by using Equation 2. The estimated set of parameters at the upper and lower boundaries can also be used to investigate how much variation in each state variable is maintained under repeated idle conditions and gas output operating conditions.

5 and 6, it is shown that the pump performance test performed for 1000 seconds consists of four steps, two separate gas output operating conditions and two idle conditions.

Table 1 shows the operating characteristic values obtained from the four state variables (suction pressure, exhaust pressure, current signals supplied to the booster pump and the dry pump). BP and DP mean booster pump and dry pump, where a _U and b _u are The slope and initial value of the upper boundary asymptote, a _L And b _L means the slope and initial value of the lower boundary asymptote.

It is evident from Table 1 that the characteristics of each state variable are explained by eight parameters. The eight parameters are two time stamps (initial and last time), four model parameters for the upper and lower asymptotes (a pair of slope and initial value), and mean and peak values. Each parameter tells when an idle or gas output condition occurs and how the gas suction pressure changes between the upper and lower boundaries in the idle or gas output state. The average value is used to check that each state variable maintains its intended level under stationary operation. Peak values are used to examine how much unexpected variation each state variable has in each idle condition and gas output condition. The transient between the idle and gas output states is broken down in the time intervals (ie, the third line of Table 1) and is not used for parameter estimation. The six operating dependent characteristic values are obtained separately for each idle state and gas output operating state (ie steps 1 to 4 in Table 1). The characteristic value is practically used to investigate how much difference the current state variable has from the initial operating conditions. Since the performance test results in Table 1 are from freshly installed vacuum pumps ready to discharge the gas produced, these results are considered the 'reference' values used for the diagnosis of the vacuum pump. Repeated performance tests under the same gas output conditions enable more stable estimates of test results when statistically judged, and as a normal performance test procedure for vacuum pumps, repeated performance tests under the same gas output conditions are recommended.

It is shown in the present invention that the operating characteristic values are obtained for two separate idle and gas output operating regions. As shown in Figs. 5 and 6, the four state variables, suction pressure, exhaust pressure and the upper and lower boundary levels of the supply current of the booster pump and the dry pump, are clearly dependent on the gas output conditions. Such mechanical and electrical state variables are generally considered to be a class of static characteristics. Unlike such static state characteristics, mechanical vibration and noise signals containing high frequency components have been used as state variables for trend and diagnostic analysis.

7 (a) to 7 (d) show vibration acceleration and noise signals and rms levels of their upper and lower boundary curves (bold solid lines), and the circular display shows each gas output condition or idle condition. Means the peak value.

7 (a) and 7 (c) show the vibration acceleration of the booster pump and the rms value of noise measured near the intermediate position between the booster pump and the dry pump. The frequency bandwidth of the vibration acceleration was selected from 10 Hz to 10 kHz, and the frequency bandwidth of the noise signal was selected from 20 Hz to 20 kHz. Both signals were digitally sampled at a rate of 40.96 kHz (ie 40,960 samples per second). Each block of 4096 samples (which corresponds to an interval of 100 ms) was used to calculate rms values shown in FIGS. 7 (a) and 7 (c). Each record of 200 sequential samples (which corresponds to a 20 second record) was used to classify the mean, maximum and minimum values of the vibration acceleration and noise level (thin solid line) shown in FIGS. 7B and 7D. The estimated model parameters of the upper and lower asymptotes and mean and peak values of the vibration acceleration and noise level are arranged in Table 2. Such model parameters were estimated from four operating conditions (two gas output operating conditions and two idle states previously given in Table 1). The vibration acceleration and noise level of FIG. 7 do not appear to exhibit noticeable gas output dependent characteristics unlike the levels of suction pressure, exhaust pressure and supply currents of the booster pump and the dry pump shown in FIGS. 5 and 6. Lose. The lower asymptotic boundary of vibration acceleration and noise level appears almost independent of the gas output condition, but the upper asymptotic boundary appears to gradually increase or decrease (ie, the slope of the positive or negative sign). Significantly varying signal elements are observed from the upper boundary level as shown in FIG. 7. The peak values of the elements fluctuating in the four operating stages, indicated by the circular symbols in FIGS. 7 (b) and 7 (d), were chosen as further operating characteristic values. The level values given in Table 2 are useful for determining how excessive vibration or noise levels have occurred during each operating phase.

Table 2 shows the operating characteristic values obtained for vibration acceleration and noise level. a _U and bU represent the slope and initial value of the upper boundary curve, and a _L and b _L represent the slope and the initial value of the lower boundary curve.

In the present invention, the results of the characterization of the state variables related to the pump operation such as the purge gas pressure and the gas flow rate, the body temperature, the coolant temperature, the lubricating oil pressure and the level of the low vacuum pump are not presented. Since the working characteristics of the state variables, which are well known in conventional trend observation and diagnostic analysis, are statistically too static, their abnormal conditions can be easily detected by using secondary statistical analysis (ie mean and standard deviation analysis). This is why the present invention does not show the analysis result of the state variable. In addition, no defective operating conditions caused by the above state variables have been observed in many cases with degraded or failed vacuum pumps. This does not mean that the operating characteristic values of the state variable are not necessary for the maintenance of the forecast of the vacuum pump. Indeed, as shown in Tables 1 and 2, obtaining the characteristic values is made for the precision diagnostic analysis proposed in the present invention.

It is interesting to note that the work characteristic values obtained for each state variable can dramatically reduce the size of the data used to perform a precise diagnosis for forecast maintenance. The current version of the electronic diagnostic guidelines [Ref. 18: Harvey Wohlwend, e-Diagnostics Guidebook, International SEMATECH, Version 1.5, October, 2002] has a minimum sampling rate of 10 Hz (10 samples per second) or higher for each state variable. High is recommended. In the present invention, the sampling rate was selected at 10 Hz according to the electronic diagnostic guidelines. As mentioned previously, a signal sampled for 1000 seconds was adopted in the present invention. The total number of samples for each state variable is estimated to correspond to 10,000. In contrast, the task characteristic values and their time stamps obtained for each static state variable were found to be only 32 data sets (four sets of six characteristic values and four sets of initial and last time stamps). . This analysis data reduction rate is extremely high. This makes it possible to greatly improve the performance of the implementation system for maintaining the forecast of the vacuum pump. Moreover, this makes it possible for each transition system to cover more vacuum pumps.

2. On-site method for obtaining performance indicators of vacuum pumps

It has been shown that operating characteristic values are obtained for individual idle conditions and gas output operating conditions. Rational measures for separating pump operating conditions are introduced in this section. Since the suction pressure was measured directly in the present invention, it is natural to use the suction pressure for such separation. When the pressure-regulated nitrogen gas is supplied to the reaction chamber, the suction pressure of the vacuum pump maintains the intended level as shown in Fig. 5 (a). This condition is referred to as the gas output condition of the vacuum pump under tests where the suction pressure level depends on the semiconductor fabrication process. After the intended time interval, the exhaust throttle valve of the reaction chamber suddenly closes almost completely. Due to the pumping operation of the vacuum pump, the suction pressure level is reduced to the basic level. The required base level also depends on the manufacturing process. This test procedure is very similar to the pump down test method for obtaining the pumping speed. If multiple gas output pressure levels are needed for the intended fabrication process, the gas output conditions corresponding to these levels can be simulated on the vacuum pump under test. As shown in Fig. 5 (a), two different gas output conditions are contemplated in the present invention, since two gas output conditions are sufficient to simulate the actual gas output conditions required for the fabrication process. The gas output simulation test proposed in this subsection makes it possible to obtain on-site pumping speed indicators which will be proved later, and this is a differential achievement contributed by the present invention. This attempt was not reported in previous inventions [Ref.1-Ref.4, Ref.6-Ref.9].

Two kinds of suction pressure transition zones, positive-going zones or negative-going zones, were observed in FIG. 5 (a). Starting a performance test of the newly installed vacuum pump, a positive-going transition occurs when the exhaust valve of the reaction chamber is opened, and a negative progression when the exhaust valve is almost closed after the gas output test phase is over. -going transition) occurs. The present invention utilizes the suction pressure measured in the secondary advancing transition region similar to the pump down test scheme.

FIG. 8 shows the suction pressure signals of the first (5 (a)) and second (5 (b)) transition zones existing between the first gas output operating region and the second idle region of FIG. 5 (a). do. Note that the bold solid line points to a suitable model of exponentially decaying function.

In the disclosure of the present invention, it has been clarified that the localization characteristic of the suction pressure signal shown in FIG. 8 is directly related to the pumping speed of the installed vacuum pump. The basic relationship between pumping speed and pump-down times, well known in vacuum theory (Ref. 20: Nigel. S. Harris, Modern Vacuum Practice, McGraw-Hill Book Company, Lendon: England, 1989), was utilized in the present invention. The algebraic equation is

In Equation 4, the symbols Q and V mean pumping speed [m ³ / h] and volume [m ³ ], respectively. The symbol ΔT means the second sampling period (ΔT = 100 [ms] in the present invention). The symbol alpha in Equation 4 is an exponential decaying constant whose value is directly related to the pumping speed.

The formula in Equation 4 is the initial value P ₀ and the last level P _n of the pressure range. It is assumed that the pumping speed is constant. As a result, by adopting the linear region on the semi-log plots shown in Fig. 8, a range suitable for the initial and final suction pressure levels is determined. The thick solid line indicates the selected area for the initial position and the final position used to estimate the exponential decay constant in the adopted pressure region. In the first zone, the initial and final pressure levels were chosen to be 80% and 20% of the observed suction pressure, respectively, before the continuous gas flow stopped. Guidelines for selecting initial and final pressure positions have proven to be very stable and efficient in estimating exponential decay constants from many field tests. Whenever necessary performance tests are required for forecasting a vacuum pump operating at the manufacturing process site, the estimated exponential decay constant is used to investigate how the pumping speed performance has decreased.

It is very straightforward to estimate the optimized exponential decay constant corresponding to the selected area. Let {Pn: n = 1, ..., N} be the suction pressure signal sampled in the selected area. The logarithm of the suction pressure signal is obtained as follows.

The optimized parameters alpha and beta are estimated by using the least-squares method described in Equation 2 and given in the previous section. The exponential decay constant optimized for the selected area is used to approximate the pumping speed indicator, which is defined as the pumping speed per unit volume, which is as follows.

The empty volume V [m ³ ] depends on the pipeline connecting the inlet of the pump and the outlet of the reaction chamber. Since this value must be a constant, it does not have to be a known value. This is the reason for using the pumping speed indicator in the present invention. Table 3 shows the estimated exponential decay constants and the estimated pumping speed indicators corresponding to the constants for the three consecutive negative transition regions of suction pressure shown in FIG. 5 (a). Note that the symbols alpha and Ip represent exponential decay constants and pumping rate indicators, respectively.

By using the measured suction pressure signal, the in-situ method proposed in the present invention to obtain the pumping speed index can be determined by the pump maintenance engineer to determine whether the target vacuum pump should be replaced by determining how much the pumping speed has been reduced. The field approach is very important as it provides the appropriate information to make the decision. The proposed field method of estimating pumping speed indicators is a very unique method that has not yet been introduced in recent pump diagnostic techniques [Ref.1-Ref.4, Ref.6-Ref.9].

The second performance indicator is related to the relationship between the difference between the mean value of the suction pressure and the mean value of the exhaust pressure, measured from the gas output and idle conditions. As shown in Table 1, the average inlet pressure differential value between the first gas output state (step 1 in Table 1) and the first idle state (step 2 in Table 1) is shown to be 9.44 [mbar]. Another mean suction pressure difference between the second gas output state (step 3 in Table 1) and the second idle state (step 4 in Table 1) is shown to be 13.4 [mbar]. In a similar manner, two exhaust pressure difference values, each with values of 4.6 [mbar] and 7.1 [mbar], are also obtained at two separate gas output operating conditions and at idle operating conditions, respectively. The ratio of intake pressure difference to exhaust pressure difference was found to be 0.49 (= 4.6 [mbar] /9.4 [mbar]) and 0.53 (= 7.1 / 13.4 [mbar]). The ratio obtained as described above is proposed in the present invention to investigate how well the relationship between the suction pressure and the exhaust pressure remains well unchanged under similar gas output conditions. Of course, reduced ratio values usually indicate a decrease in pump performance.

It should be noted here that the supply current to the booster pump is also closely related to the gas output level. The average supply current difference between the first gas output state (step 1 in table 1) and the first idle state (step 2 in table 1) is shown to be 2.71 [A]. Another mean supply current difference between the second gas output state (step 3 in Table 1) and the second idle state (step 4 in Table 1) is observed to be 3.88 [A]. The ratio of the suction pressure difference value to the supply current difference value is 0.29 [A / mbar] (= 2.71 [A] /9.44 [mbar]) and 0.29 [A / mbar] (= 3.88 [A] /13.4 [mbar]) Was found to be. The two ratio values appear to be approximately equal to each other. This is a very useful characteristic for investigating how much current is supplied to the booster pump compared to the initial operating conditions. Indeed, the ratio of the suction pressure difference value to the supply current difference value is the third pump performance index proposed in the present invention. In a manner similar to the way of dealing with the current supplied to the booster pump, two difference values of the average current value supplied to the dry pump are obtained. The average supply current difference between the first idle state (step 2 in Table 1) and the first gas output state (step 1 in Table 1) is shown to be 0.3 [A]. Another average supply current difference between the second idle state (step 4 in Table 1) and the second gas output state (step 3 in Table 1) is shown to be 0.4 [A]. The ratio of the suction pressure difference value to the supply current difference value is 0.032 [A / mbar] (= 0.3 [A] /9.44 [mbar]) and 0.030 [A / mbar] (= 0.4 [A] /13.4 [mbar]), respectively. Was found to be. The dependence of the current supplied to the dry pump on the suction pressure is the fourth vacuum pump performance index proposed in the present invention. In detail, the third and fourth performance indicators are the mechanical loads applied to the booster pump and the dry pump's electric motor, because the supply currents for the booster pump and the dry pump are directly related to the mechanical loads on the motors of both pumps. It should be noted that enabling the analysis of (torque) conditions.

Four types of vacuum pump performance indicators and how to obtain them are described in detail in this section. The obtained results of the performance indicators are shown, which are actually considered standard (or initial) performance data for a newly installed vacuum pump. The manner in which the above performance indicators are used to investigate how the current performance of a working vacuum pump differs from standard operating data will be presented in the following subsection.

3. Precise diagnosis process to maintain forecasts of degraded vacuum pumps

In the previous two subsections, an active algorithm of working characteristics for each measured state variable and how to obtain four vacuum pump performance indicators in the field are described in detail. When the active algorithm and the developed field method were applied to the initial performance test of newly installed low vacuum pumps, they were found to provide estimated characteristic values and obtained performance indicators. In this subclause, the possibility of such a precision analysis method is applied by applying a precision analysis to maintain the forecast of the vacuum pump to a vacuum pump that has reached abnormal operating conditions (i.e. 'slowly reacting' to the intended vacuum level of the reaction chamber). Efficiency is investigated. This unforeseen vacuum pump operating condition is encountered after numerous normal operations. The first thing that maintenance engineers need to do to maintain the forecast of the target vacuum pump is to perform on-site (or outdoor) performance tests according to a predetermined test procedure, and the initial (or standard) performance of the current performance test results stored in the maintenance database of the vacuum pump. It is to determine how much difference there is from the value. However, this theoretical approach to forecasting vacuum pumps has not been well established among major semiconductor manufacturers or vacuum pump suppliers.

The estimated working characteristic values and the performance indicators obtained in the previous section are actually obtained from the first stage of the target vacuum pump operation, as mentioned previously. The results presented above are referred to as reference data in the present invention. 9 shows time signals measured from a vacuum pump operating under abnormal operating conditions. The signal is six state variables, 6 (a) is the suction pressure signal, 6 (b) is the exhaust pressure signal, 6 (c) is the current supplied to the booster pump, 6 (d) is the current supplied to the dry pump, 6 (e) is a mechanical vibration signal, 6 (f) is a noise signal, and their respective reference characteristic values are already shown in Tables 1 and 2. As shown by the suction pressure signal in FIG. 9 (a), the pump operation in two simple steps (gas output and idle) was investigated. The first stage is an operating condition where the gas output is gradually increased and the second stage is an idle state where the suction pressure maintains the basic level. The two working conditions are separated by a short state transition region between the 290 second position and the 320 second position. It is not easy to distinguish any notable abnormal signal element from the state variable signal shown in FIG. Initial signals corresponding to six state variables, considered as reference values, are shown in FIGS. 5 (a) to 5 (d) and 7 (a) and 7 (c), which are newly installed vacuum pumps. It was measured at the early stage of. By comparing the signals, it may be observed that the mechanical vibration and noise levels of the current vacuum pump have increased further than the levels at the initial performance test conditions. These observed results are not sufficient for maintenance engineers to determine whether the target vacuum pump should be replaced. This is one of the most challenging issues many repair engineers face in real semiconductor fabrication.

Given the measured signals under two operating conditions dependent on gas output, the first diagnostic step is to estimate the working characteristic values for each state variable in the same way as was done in section 1. Table 4 shows a list of estimated work characteristic values of the six measured state variables shown in FIG. 9. BP and DP mean booster pump and dry pump, note that a _U and b _U mean the slope and initial value of upper boundary asymptote, and a _L and b _L mean the slope and initial value of lower boundary asymptote. do.

In gas output operating conditions, the average suction pressure level is 20.3 [mbar] and the default suction pressure is 0.73 [mbar]. These gas output operating ranges are shown for the third and fourth operating stages with initial gas output operating conditions, i.e. in Table 1, with an average suction pressure of 14 [mbar] for each gas output operating condition and suction pressure of 0.7 [mbar] for idle conditions. Pretty close to the second case. This small change in suction pressure at the gas output is not due to the deterioration of the vacuum pump, but to the process conditions associated with the final product. Moreover, the average exhaust pressure at idle is the initial value of 1000.3 [mbar] (ie the mean value of the exhaust pressure in steps 2 and 4 of Table 1) and the current value of 1007.3 [mbar] (ie the mean value of exhaust pressure in step 2 of Table 4). This is due to the increase in atmospheric pressure with dates. In order to overcome the effect of unwanted atmospheric pressure on these measured exhaust pressure levels, the present invention proposes the use of the relative value of the exhaust pressure to the average exhaust pressure level in the idle state, as mentioned in Section 2. The exhaust pressure level difference value in the gas output state and the idle state is used in the present invention. The exhaust pressure differential values for both cases in the initial working state (ie steps 3 and 4 in table 1) and in both cases in the current state (steps 1 and 2 in table 4) are 7.1 [mbar] and 7.6 [mbar], respectively. Observed. Small changes in the average suction pressure of the gas output state from 14 [mbar] to 20.3 [mbar] do not cause a noticeable increase in exhaust pressure. This may indicate that the exhaust pressure signal is not sensitive to changes in the gas output level of the vacuum pump. This is why the present invention strongly recommends the direct use of the suction pressure signal for state observation and diagnostic analysis for the maintenance of the forecast of the vacuum pump.

In Table 4, the average value of the booster pump (shortened to BP) supply current in the current gas output operating state is 8.93 [A]. This is slightly increased compared to the initial average value of 8.22 [A] (average BP supply current value in step 3 of Table 1). In the current idle operation, the average value of BP supply current is observed to be 3.86 [A], which is slightly lower than the initial average value of 4.34 [A] (average BP supply current value in step 4 of Table 1). Unlike these small changes in BP supply current, the average value of the dry pump (reduced by DP) supply current measured at gas output and idle is 15.8 [A] and 16.5 [A], respectively, which is the initial value, i.e. It appears to be close to 15.8 [A] and 16.2 [A] (average DP supply current values in steps 3 and 4 of Table 1). This average supply current value, measured from the current abnormal operating conditions of the target vacuum pump, shows no noticeable change at all sufficient to determine the performance degradation of the pump.

It is shown in Table 4 that the average vibration level measured in the current dry pump increased from the initial level value of 10.8 [m / s ² , rms] to 13.4 [m / s ² , rms] (steps 3 and 3 of Table 2). 4 levels of vibration are standard). The peak value of the acceleration level corresponding to the vibration level average value also increased from an initial value of 12.9 [m / s ² ] to a current value of 16.7 [m / s ² ]. Similarly, the current average noise level increased from the initial value of 0.85 [Pa, rms] to 1.07 [Pa, rms] (standard noise levels in steps 3 and 4 of Table 2). The peak value corresponding to the average noise level was also increased from the initial value of 1.32 [Pa] to the current value of 2.63 [Pa]. These results indicate that the current operating conditions of the vacuum pump are 24%, 29% rise in mean-squared mean and peak values for mechanical vibration, and mean-squared mean and peak values for noise levels. This indicates that the increase was 26% and 99%. Mechanical vibration and noise signals, which are considered to be dynamic state variables, have been shown to yield relatively increased evaluation results for the current operating conditions than the four static state variables mentioned above. The two dynamic state variables are more sensitive to changes in the operating conditions of the vacuum pump than static variables. However, since the increased level values observed from the mechanical vibration and noise signals are considered acceptable when compared to the diagnostic guidelines related to mechanical vibration [Ref. 2], this increase is determined by the maintenance engineer. It is not enough to decide whether to replace the target vacuum pump with a new one.

As a result, the existing diagnostic approaches discussed above have been found to be unsuccessful in diagnosing possible causes under abnormal operating conditions of the target vacuum pump. This may indicate that existing diagnostic approaches using level analysis of measured state variables provide only limited capability for precision diagnostics to detect the degraded performance of low vacuum pumps early. This limited capability of the existing diagnostic approach provided the main motivation for developing the direct vacuum pump performance indicators proposed in Section 2. The first performance factor proposed in the present invention is the pumping speed index obtained from the measured suction pressure signal.

10 shows the suction pressure signal measured from the sub advancing transition region. The thick line indicates the values obtained from the optimized model used to obtain the puncturing velocity index. The pumping rate indicators obtained are shown in Table 5.

Current pumping speed indicators, derived from abnormal operating conditions, show a sharp decrease by 31% compared to the initial operating conditions. This 61% reduction in pumping speed was clearly found to result in a slow reaction to the intended vacuum level, i.e. the abnormal operating conditions of the subject vacuum pump. The proposed pumping speed indicator proved to be very efficient for the precise diagnosis of vacuum pumps. Such a reduction in pumping speed may indicate that the target pump should be replaced with a new one as soon as possible. It is interesting to note that although the pumping speed has been reduced by 69%, the ratio of the inlet pressure difference value to the exhaust pressure difference value has only been reduced by 26%. This means that even if the average exhaust pressure is not so sensitive to the drop in pumping speed, the average exhaust pressure value still provides useful information about its performance degradation. The reason is that the pumping speed is almost related to the rate of decrease of the suction pressure, not the constant state basic level of the suction pressure and the exhaust pressure. Unlike these two performance indicators, no noticeable change is observed from the supply current related performance indicators given in Table 5. This may indicate that no external mechanical loads (torques) on the booster and dry pumps have been increased. The use of the suction pressure measured as the state variable of the vacuum pump not only enabled the precise diagnosis of the low vacuum pump but also led to the stable forecast of the vacuum pump. This shows how important the suction pressure signal is for the precise analysis for forecast maintenance. This is one of the major contributions of the present invention.

The six state variables considered in this subclause are suction and exhaust pressures, current supplied to booster pumps and dry pumps, mechanical vibration and noise signals. The active algorithm for estimating the model parameters suitable for the operating conditions of the vacuum pump is optimized model parameters {a _U , b _U , a _L , b _L } and mean values for the gas output operating conditions at each idle and gas output state. And peak values. As a result, the six parameters are representative data sets for each operating condition (idle state or gas output state). Whenever a performance test is needed, a series of six parameters for all considered state variables appear as a pair of two-dimensional matrices.

Note that the subscript symbols ("IDLE" and "LOAD") mean idle operating conditions and gas output operating conditions. The raw column index n represents a sequence of performance tests. The column indexes j and k denote a classification number of seven state variables and an order of six parameters for each state variable. Although the mechanical vibration signal measurement of the booster pump is not disclosed in the present invention, the seventh state variable corresponds to the vibration signal measurement. If necessary, a pumping speed indicator may be included in the last column of the matrix. The selection of the classification number and the parameter order can be done in any convenient way. When either an idle state or a gas output operating state is transitioned, a horizontal column vector corresponding to it is obtained. As the idle operation and the gas output operation are repeated, two matrices are obtained.

Statistical analysis of single or multiple variables, Mahana nobis distance analysis [Ref.10] and process performance analysis [Ref.21: Z.G. Stoumbos, "Proess capability indices: Review and extensions," Nonlinear Analysis: Real World Applications, Vol. 3, pp. By utilizing well-known analysis methods such as 191-210, 2002, the matrix data described in Equation 7 can be easily used for precise analysis of the target vacuum pump. Indeed, in this subclause, the concept behind the statistical analysis of a single variable is used to indicate how well the estimated characteristic values can be used for precision diagnostic analysis to maintain forecasts of low vacuum pumps. The concepts and logical sequence disclosed in the previous subclauses fit well with the statistical analysis of single variables. However, multivariate analysis, including process performance analysis and Mahana nobis distance analysis, has not yet been considered, since the technical discussion thereof is beyond the scope of the present invention. The present invention prefers Mahana nobis distance analysis over multivariate analysis or process performance analysis. Because the Mahana Nobis distance analysis gives us a more sensitive response to small changes in the estimated model parameters. Matrix data consisting of optimized model parameters and average peak values of dynamic characteristic values imparted to the measured state variables lead to another efficient means for precise analysis for predictive maintenance of vacuum pumps. The conversion of structured data matrices into existing analytical algorithms such as multivariate statistical analysis, process performance analysis, and Mahana nobis distance analysis is, of course, one of the contributions contributed by the present invention.

Fulfillment system

This section introduces the technical details necessary to realize a system for implementing a precision diagnostic analysis for forecasting low vacuum pumps. FIG. 11 shows a schematic arrangement of state variables measured by a system of implementation of a precision diagnostic analysis for maintenance of low vacuum pumps. As already introduced in the previous section, the transition system has seven state variables (suction and exhaust pressures, currents supplied to the booster and dry pumps, mechanical vibration signals of the booster and dry pumps and between the booster and dry pumps). Noise signal measured from the gap).

In order to measure the suction pressure, an inlet pressure gauge (or vacuum gauge) capable of measuring the appropriate range for the multiple gas output conditions is installed at the inlet of the vacuum pump. In order to measure the exhaust pressure, an exhaust pressure transducer capable of measuring a full range of two to three times the atmospheric pressure level is installed in the vicinity of the exhaust port of the dry pump. In order to measure the supply current of the booster pump and the dry pump, a current probe is used that can withstand the peak value of the current level. To measure the mechanical vibrations of the booster pump and the dry pump, two accelerometers capable of withstanding the high temperature range (i.e. 150 ° C or higher) are mounted fixed to the body surface of the booster pump and the dry pump. To measure the noise from the vacuum pump, a measurement microphone dedicated to the high sound pressure range (1/4 "size model) is installed in the middle of the booster pump and the vacuum pump. These mechanical and electrical sensors are dedicated signals Connected to a control unit, the signal control unit not only provides each sensor dependent input voltage, but also amplifies the sensor output voltage to external equipment up to the intended level. Recommends the use of the mechanical and electrical sensors listed above for the implementation of precision diagnostics for failure prevention and predictive maintenance of low vacuum pumps used in modern semiconductor fabrication processes. Not all of them use one or more sensor outputs listed above for external use. If possible, the output signals of the sensors may also be coupled to a conserved input channel of the signal adjustment unit.

Other state variables such as purge gas pressure, purge gas flow rate, body temperature, coolant temperature, lubricant pressure and level of vacuum pump are already available from existing pump observation systems already installed in low vacuum pumps, so they are not measured directly. Did. Indirect collection of the state variables is achieved with the aid of PC-based data reading software provided by the pump supplier. When an external connection to the state variables collected indirectly is available, a multi-channel data acquisition (DAQ) system of the transition system is constructed to read the signal of the state variables.

Fig. 12 shows a system for implementing a precise diagnosis analysis for maintaining the forecast of a vacuum pump installed in a semiconductor manufacturing company. It consists of three main parts: signal processing units, multichannel data acquisition (DAQ) systems, and server class PC systems. The signal processing unit dedicated to the seven sensors introduced above includes a unique input power supply and output voltage amplifier for each sensor only. High-performance DAQ systems consist of any 16-channel samples and holders with sampling rates up to 100kHz and 16-channel 24-bit analog-to-digital (AD) converters. A dual processed server class PC system was specifically selected for stable processing of DAQ system sample data through the interface of MXI-II provided by National Instrument Co. In the present invention, a data acquisition program was developed by using a LabView program provided by National Instruments. The PC system also implements all the digital signal processing schemes already mentioned in the previous section. Smaller versions of the transition system are being developed that are portable and mobile at the actual process site. The size of this version is expected to be much smaller than the current implementation system shown in FIG.

According to the present invention, the most challenging issue in the present invention has been to find a systematic method that allows maintenance engineers to determine the appropriate time for replacement of a vacuum pump based on their current performance test test results. In addition, the comparison of current diagnostic data with initial (or standard) data sets has been shown to allow a maintenance engineer to determine whether to replace a target vacuum pump based on derived pump performance indicators. In addition, the derived pumping speed indicators of current vacuum pumps are reported to be generally decreasing. These quantitative diagnostic analysis results are expected not only to determine the appropriate time for replacement of the vacuum pump based on their current performance performance results, but also to improve the stability and reliability of forecast maintenance of low vacuum pumps.

Claims (11)

1) collecting state variables related to various pump operations of the newly installed vacuum pump at preset sampling rates for idle conditions and various gas output conditions; 2) the maximum of the time series values of the state variables from each set of successively sampled signals, in a period of user-adopted periods longer than the dominant period of the state variable signal element that fluctuates under idle conditions and various gas output conditions. Determining a minimum value; 3) estimating pump operating characteristic values using an active diagnostic algorithm based on parametric models under idle conditions and various gas output conditions; 4) obtaining a pump performance index of the newly installed vacuum pump using an in-situ evaluation method; 5) storing estimated pump operating characteristic values and pump performance indicators of the newly installed vacuum pump in a vacuum pump maintenance database; 6) repeating steps 1 to 5 whenever the newly installed vacuum pump is observed to be in an abnormal operating condition; 7) Estimated pump operating characteristic value and pump performance index of the newly installed vacuum pump stored in the vacuum pump maintenance database to determine whether to replace the pump, estimated pump operating characteristic value and pump performance index under abnormal operating conditions Precision diagnostic method for failure prevention and forecast maintenance of the vacuum pump, characterized in that consisting of a step of comparing with. The method of claim 1, The vacuum pump is a precision diagnostic method for failure prevention and forecast maintenance of the vacuum pump, characterized in that used in the semiconductor manufacturing process. The method of claim 1, State variables related to pump operation include fault and exhaust pressures, supply currents of booster pumps and dry pumps, rms values of mechanical vibration and noise signals. Precision diagnostics. The method of claim 1, The dominant period of the changing state variable signal element is determined by using fast fourier transfrom (FFT) analysis and is used to classify the maximum and minimum values among successively sampled time series values of the state variables. Diagnostic method to prevent failures and maintain forecasts of vacuum pumps. The method of claim 1, The pump operating characteristic values may include optimized model parameters, an idle condition, an average value and a peak value of each state variable measured under a gas output (load) condition. Precise diagnostics to maintain forecasts. The method of claim 5, The optimized model parameters are calculated by using Equation 1 below, which is a linear model that can represent the upper and lower asymptotes of each measured state variable. (Equation 1)

In Equation 1, the subscript k denotes an upper or lower asymptotic model {k = U or L}, and the time series values of the maximum and minimum values classified in each pumping state are {y _{k, n} : n = 1,2 , ..., N}, the optimized model parameters are obtained by using the least squared method of Equation 2 below. (Equation 2)

In Equation 2, the first parameter {a _k : k = U or L} is the slope of the upper and lower asymptotes of each measured state variable, and the second parameter {b _k : k = U or L} is the upper and lower asymptotes. Refers to each initial value of. The method of claim 1, The pump performance index may include a pumping speed index obtained by an in-situ evaluation method, which is obtained from the pump operating characteristic values; Ratio of suction pressure difference value to exhaust pressure difference value; Ratio of suction pressure difference value to supply current difference value of BP (booster pump); And a ratio of the difference between the suction pressure difference value and the supply current difference value of the DP (dry pump). The method of claim 7, wherein The pumping speed index is a precision diagnostic method for preventing failure and maintaining the forecast of the vacuum pump, characterized in that defined by the pumping speed per volume as shown in Equation 3 below. (Equation 3)

In Equation 3, the symbols alpha and Ip mean exponentially decaying constants and pumping speed indices, and the symbols Q and V mean pumping speeds [m ³ / h] and empty volumes [m ³ ], The exponential decay constant is a logarithm value of a suction pressure signal {Pn: n = 1, ..., N} corresponding to a negative going transition region defined by Equation 4 below. Obtained from. (Equation 4)

The estimation of the exponential decrement constant alpha and initial value optimized in Equation 4 is obtained by using a least squared method. The method of claim 1, The in-situ evaluation method includes a method in which a negative going transition region of suction pressure is used to obtain a pumping speed, and a suction pressure signal is measured as a reference. Precision diagnostic method for failure prevention and forecast maintenance of the vacuum pump, characterized in that. The method of claim 1, The precision diagnosis method includes a design of a pair of two-dimensional structure matrix including estimated values of pump operating characteristics of state variables related to pump operation obtained at each idle condition and gas output condition as shown in the following equation. Precise diagnosis to prevent breakdown of vacuum pumps and to maintain forecasts.

for n = 1, ...; j = 1, ..., J (total number of state variables related to measured pump operation) k = 1, ..., K (total number of pump operating characteristic values) In the above equations, the subscripts "IDLE" and "LOAD" denote idle conditions and gas output operating conditions, horizontal index n denotes a sequence of performance tests, and vertical indexes j and k represent classifications of state variables related to pump operation. Number and sequence of pump operating characteristic values for each state variable. A signal control unit for amplifying an inlet pressure gauge, an exhaust pressure transducer, a booster pump, a supply current detector of a dry pump, a mechanical vibration sensor, and a measurement microphone centrally installed in the booster pump and the dry pump; A high speed multichannel data acquisition (DAQ) system suitable for collecting mechanical vibration and sound pressure signals with ultrahigh frequency components ranging from 10 kHz to 20 kHz; And It consists of a dual processor server-class PC system that performs the following steps: The PC system collects all state variable values measured from a real-time DAQ system without losing any data transmission, calculates the root mean square values of mechanical vibration and sound pressure signals, and continuously operates at idle and other gas output conditions. The maximum and minimum time series values of the measured state variable values are determined from each group of sampled signals, and the average and maximum values of the optimized model parameters and the measured state variables are determined using an active diagnostic algorithm based on a parametric model. Estimate the pump automatic characteristic values, obtain the pump performance index of the newly installed vacuum pump by using the in-situ evaluation method, and estimate the estimated pump operating characteristics and pump performance of the newly installed vacuum pump. The indicators are stored in the vacuum pump maintenance database. For the sake of brevity, the estimated pump operating characteristic value and pump performance index of the newly installed vacuum pump stored in the vacuum pump maintenance database are compared with the estimated pump operating characteristic value and pump performance index under abnormal operating conditions. Precise diagnosis system for failure prevention and forecast maintenance of vacuum pump.

Images