A fault prediction method for vacuum dry pump based on distribution feature subset

Abstract

The present invention discloses a vacuum dry pump fault prediction method based on distribution feature subsets, comprising: constructing a vacuum dry pump operation state vector set DPVS; constructing a data distribution vector set DPFI based on feature subsets; constructing a vacuum dry pump distribution feature subset DPFS; calculating the distance between vacuum dry pump distribution feature subsets according to DPFS to obtain a vacuum dry pump distribution feature subset distance list fs_dist to be predicted; constructing a dry pump fault prediction training set FDDPD based on distribution feature subsets according to DPFI and fs_dist; and establishing a vacuum dry pump fault prediction model BFPM for FDDPD using a Logistic classifier. The present invention achieves a better vacuum dry pump fault prediction effect.

Description

Technical Field

[0001] The invention relates to a vacuum dry pump fault prediction method based on a distribution feature subset. Background Art

[0002] With the rapid development of electronic technology, the semiconductor industry, as the core of the modern electronics industry, has put forward higher requirements for high-precision and high-efficiency production processes. Vacuum technology and its core components play a key role in semiconductor manufacturing. Vacuum dry pumps are equipment for creating vacuum environments and play a vital role. Therefore, more and more researchers are investing in the research field of vacuum dry pump failure prediction. They are committed to building accurate dry pump failure prediction models and using these models and tools to predict dry pump failures to ensure the safety of semiconductor production.

[0003] The operating data of faulty equipment is very important for building an accurate dry pump fault prediction model. Therefore, researchers need to fully explore the abnormal information in the downtime data and select more samples from the downtime data as reasonably as possible based on the operating data at the prediction time.

[0004] In order to further optimize the sample selection of the downtime history database, a vacuum dry pump fault prediction method based on distribution feature subsets was provided. Summary of the invention

[0005] The purpose of the present invention is to overcome the existing defects and provide a vacuum dry pump fault prediction method based on a distribution feature subset, thereby achieving a better vacuum dry pump fault prediction effect.

[0006] The technical solution to achieve the above purpose is:

[0007] A vacuum dry pump fault prediction method based on a distribution feature subset, comprising:

[0008] Step S1, constructing a vacuum dry pump operation state vector set DPVS;

[0009] Step S2, constructing a data distribution vector set DPFI based on a feature subset according to the vacuum dry pump operation state vector set DPVS;

[0010] Step S3, constructing a vacuum dry pump distribution feature subset DPFS according to the vacuum dry pump operation state vector set DPVS and the data distribution vector set DPFI based on the feature subset;

[0011] Step S4, calculating the distances between the vacuum dry pump distribution feature subsets according to the vacuum dry pump distribution feature subset DPFS, and obtaining a distance list fs_dist of the vacuum dry pump distribution feature subsets to be predicted;

[0012] Step S5, constructing a dry pump fault prediction training set FDDPD based on a distribution feature subset according to the data distribution vector set DPFI based on the feature subset and the distance list fs_dist of the vacuum dry pump distribution feature subset to be predicted;

[0013] Step S6, using a Logistic classifier to establish a vacuum dry pump fault prediction model BFPM for the dry pump fault prediction training set FDDPD based on the distribution feature subset, and using the vacuum dry pump fault prediction model BFPM to perform vacuum dry pump fault prediction on other downtime pump data.

[0014] Preferably, the step S1 comprises:

[0015] Step S11, obtaining a set of operating data of vacuum dry pumps of the same type based on actual applications of vacuum dry pumps in the market;

[0016] Step S12, constructing a vacuum dry pump operation data subset with different numbers under the same vacuum dry pump type;

[0017] Step S13, constructing a second-level vector of the vacuum dry pump operation status {Time, DryPumpPower, BoosterPower, DPInverterSpeed, MBInverterSpeed, DPTemperature, MBTemperature, N2 Flow, ExhaustPressure} based on the dry pump motor power, motor speed, sensor temperature, nitrogen flow and exhaust pressure parameters;

[0018] Step S14, the running status data set of the vacuum dry pump that has been shut down is processed by the step S13 to obtain a shut down running status vector set EDPVS = [rsv1, rsv2, …, rsv i ], where i=1,2,3,…,n;

[0019] Step S15, the operating state data set corresponding to all moments in the vacuum dry pump to be tested is processed by the step S13 to obtain the operating state vector set TDPVS of the vacuum dry pump to be tested = [rsvp1, rsvp2, …, rsvp j ], where j=1,2,3,…,m;

[0020] Step S16, constructing the downtime operation status label set EL=[el1, el2,…, el x ], where x=1,2,3,…,n; the label set corresponds to each vacuum dry pump in the downtime operation state vector set ERSVS in step S14;

[0021] Step S17, constructing a vacuum dry pump operation state vector set DPVS={EDPVS, EL, TDPVS} through the obtained downtime operation state vector set EDPVS, the vacuum dry pump operation state vector set TDPVS to be tested and the downtime operation state label set EL.

[0022] Preferably, in step S13, Time represents the current time when the equipment is running, DryPumpPower represents the working power of the lower dry pump, BoosterPower represents the working power of the upper dry pump, DPInverterSpeed ​​represents the working frequency of the lower dry pump, MBInverterSpeed ​​represents the working frequency of the upper dry pump, DPTemperature represents the working temperature of the lower dry pump, MBTemperature represents the working temperature of the upper dry pump, N2 Flow represents the nitrogen flow rate introduced into the dry pump, and ExhaustPressure represents the exhaust pressure of the dry pump.

[0023] Preferably, the step S2 comprises:

[0024] Step S21, setting the basic unit of data distribution vector processing to hours;

[0025] Step S22, reading the operating status data trsd of a single vacuum dry pump;

[0026] Step S23, the read operation status data trsd is processed by data distribution vector to obtain the unit vector set trsdb of a single dry pump, and the time corresponding to the first row of data in the unit vector set trsdb is used as the time of the trsdb, recorded as trsdb_t;

[0027] Step S24, read a running state feature sf, calculate the minimum value, maximum value, average value, standard deviation and median of the running state feature sf, and use these five values ​​as the data distribution vector of the feature, recorded as sfdv;

[0028] Step S25, traverse each operating state feature sf in the unit vector set trsdb of the single dry pump, and obtain the vacuum dry pump operating data distribution vector through the processing in step S24, which is in the form of [sfdv1, sfdv2,…, sfdv k ], where k=1,2,3,…,8;

[0029] Step S26, processing the downtime operation state vector set EDPVS in the vacuum dry pump operation state vector set DPVS in the step S17 through the steps S21 to S25 to obtain the downtime operation state data distribution vector set DPFI-E;

[0030] Step S27, processing the vacuum dry pump operation state vector set TDPVS to be tested in the vacuum dry pump operation state vector set DPVS in the step S17 through the steps S21 to S25 to obtain a vacuum dry pump operation state data distribution vector set DPFI-T to be predicted;

[0031] Step S28, marking the time corresponding to the first row of data of each processing unit in the downtime operation status data distribution vector set DPFI-E as the time of the unit distribution vector, and obtaining the downtime operation status label set EL;

[0032] Step S29, constructing the dry pump operation data distribution vector set DPFI={DPFI-E, EL, DPFI-T} through the obtained downtime operation status data distribution vector set DPFI-E, the vacuum dry pump operation status data distribution vector set DPFI-T to be predicted and the downtime operation status label set EL.

[0033] Preferably, step S3 comprises:

[0034] Step S31, obtaining the constructed second-level vector of the operating state of the vacuum dry pump, and selecting a subset svs in the second-level vector of the operating state of the vacuum dry pump;

[0035] Step S32, selecting the data distribution vector corresponding to the original feature from the downtime operation status data distribution vector set DPFI-E according to the vacuum dry pump operation status second-level vector subset svs, and obtaining the downtime history data distribution feature subset DPFS-S;

[0036] Step S33, selecting the data distribution vector corresponding to the original feature from the data distribution vector set DPFI-T of the vacuum dry pump operating state to be predicted according to the second-level vector subset svs of the vacuum dry pump operating state, and obtaining the distribution feature subset DPFS-T of the vacuum dry pump to be predicted;

[0037] Step S34, obtain the downtime operation status label set EL, and then construct the dry pump distribution feature subset DPFS={DPFS-S, EL, DPFS-T} according to the downtime history data distribution feature subset DPFS-S, the predicted vacuum dry pump distribution feature subset DPFS-T and the downtime operation status label set EL.

[0038] Preferably, step S4 comprises:

[0039] Step S41, obtaining a distribution feature subset DPFS-S of downtime history data and a distribution feature subset DPFS-T of a vacuum dry pump to be predicted;

[0040] Step S42, selecting a feature subset vector corresponding to a certain device at a certain time in the vacuum dry pump distribution feature subset DPFS-T to be predicted;

[0041] Step S43, sequentially calculate the Euclidean distance of the feature subset vector at each moment in the downtime history data distribution feature subset DPFS-S, and obtain a list of vacuum dry pump distribution feature subset distances to be predicted fs_dist = [dist1, …, dist e ], where e=1,2,3,…,m.

[0042] Preferably, step S5 comprises:

[0043] Step S51, sorting the obtained distance list fs_dist of the distribution feature subset of the vacuum dry pump to be predicted from small to large, and recording the sorted index index_fs;

[0044] Step S52, selecting the first k indexes in the sorted index index_fs, and constructing the first k most similar index set MS_k;

[0045] Step S53, traverse all subsets in the predicted vacuum dry pump feature subset DPFS-T and execute the steps S51 to S52 to obtain the most similar index set MS_k of the distribution feature subset v ;

[0046] Step S54: The most similar index set MS_k is obtained. v Combine to get the most similar repeated index set Repeated-index v ;

[0047] Step S55: The most similar repeated index set Repeated-index v Remove duplicate indexes to obtain the training set index train-index required for the dth downtime device at that moment d ;

[0048] Step S56, taking out the downtime operation status data distribution vector set DPFI-E according to the training set instance index train-indexi of step S55, and obtaining a feature set FDDPD-data based on the distribution feature subset;

[0049] Step S57, taking out the downtime operation status label set EL according to the training set instance index train-indexi of step S55, and obtaining the label set FDDPD-label based on the distribution feature subset;

[0050] Step S58, constructing a vacuum dry pump fault prediction training set FDDPD={FDDPD-data, FDDPD-label} based on the distribution feature subset according to the feature set FDDPD-data based on the distribution feature subset and the label set FDDPD-label based on the distribution feature subset.

[0051] The beneficial effects of the present invention are as follows: the vacuum dry pump fault prediction method based on the distribution feature subsets of the present invention selects different samples by dynamically adjusting the number of feature subsets and calculating the Euclidean distance between feature subsets during the sample selection process of the downtime history database, and the samples obtained under different feature subsets are merged to form a final training sample set. This method helps to achieve better vacuum dry pump fault prediction effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a flow chart of a vacuum dry pump fault prediction method based on a distribution feature subset of the present invention;

[0053] Figure 2 It is a specific flow chart of constructing the vacuum dry pump operation state vector set DPVS in the present invention;

[0054] Figure 3 It is a specific flow chart of constructing a data distribution vector set DPFI based on a feature subset in the present invention;

[0055] Figure 4 It is a specific flow chart of constructing the vacuum dry pump distribution feature subset DPFS in the present invention;

[0056] Figure 5 is a specific flow chart for calculating the distance between dry pump distribution feature subsets in the present invention;

[0057] Figure 6 It is a specific flow chart of constructing a vacuum dry pump fault prediction training set FDDPD based on a distribution feature subset in the present invention;

[0058] Figure 7 It is a schematic diagram of the number of historical distribution vectors selected under different k in the present invention;

[0059] Figure 8 It is a schematic diagram of the early warning situation obtained by using the Logistic classifier in the present invention. DETAILED DESCRIPTION

[0060] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. In the description of the present invention, it should be noted that the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside" and the like indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings, which are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0061] The present invention will be further described below in conjunction with the accompanying drawings.

[0062] like Figure 1 As shown, a vacuum dry pump fault prediction method based on a distribution feature subset includes:

[0063] Step S1, constructing a vacuum dry pump operation state vector set DPVS.

[0064] like Figure 2 As shown, step S1 specifically includes:

[0065] Step S11, obtaining a set of operating data of vacuum dry pumps of the same type based on actual applications of vacuum dry pumps in the market.

[0066] Step S12, constructing a vacuum dry pump operation data subset with different numbers under the same vacuum dry pump type.

[0067] Step S13, based on the dry pump motor power, motor speed, sensor temperature, nitrogen flow and exhaust pressure parameters, construct a second-level vector of the vacuum dry pump operation status {Time, DryPumpPower, BoosterPower, DPInverterSpeed, MBInverterSpeed, DPTemperature, MBTemperature, N2 Flow, ExhaustPressure}.

[0068] In the embodiment, Time represents the current time when the device is running, DryPumpPower represents the working power of the lower dry pump, BoosterPower represents the working power of the upper dry pump, DPInverterSpeed ​​represents the working frequency of the lower dry pump, MBInverterSpeed ​​represents the working frequency of the upper dry pump, DPTemperature represents the working temperature of the lower dry pump, MBTemperature represents the working temperature of the upper dry pump, N2 Flow represents the nitrogen flow rate introduced into the dry pump, and ExhaustPressure represents the exhaust pressure of the dry pump.

[0069] Step S14, the running status data set of the downtime vacuum dry pump is processed by step S13 to obtain the downtime running status vector set EDPVS = [rsv1, rsv2, …, rsv i ], where i=1,2,3,…,n.

[0070] Step S15, the operating status data set corresponding to all moments in the vacuum dry pump to be tested is processed by step S13 to obtain the operating status vector set TDPVS of the vacuum dry pump to be tested = [rsvp1, rsvp2, …, rsvp j ], where j=1,2,3,…,m.

[0071] Step S16, constructing the downtime operation status label set EL=[el1, el2,…, el x ], where x=1,2,3,…,n; the label set corresponds to each vacuum dry pump in the downtime operation state vector set ERSVS in step S14.

[0072] Step S17, constructing a vacuum dry pump operation state vector set DPVS={EDPVS, EL, TDPVS} through the obtained downtime operation state vector set EDPVS, the vacuum dry pump operation state vector set TDPVS to be tested and the downtime operation state label set EL.

[0073] Step S2, constructing a data distribution vector set DPFI based on a feature subset according to the vacuum dry pump operation state vector set DPVS.

[0074] like Figure 3 As shown, step S2 specifically includes:

[0075] Step S21, setting the basic unit of data distribution vector processing to hours.

[0076] Step S22, reading the operating status data trsd of a single vacuum dry pump.

[0077] Step S23, the read operation status data trsd is processed through the data distribution vector to obtain the unit vector set trsdb of the single dry pump, and the time corresponding to the first row of data in the unit vector set trsdb is used as the time of the trsdb, recorded as trsdb_t.

[0078] Step S24, read a running status feature sf, calculate the minimum value, maximum value, average value, standard deviation and median of the running status feature sf, and use these five values ​​as the data distribution vector of the feature, recorded as sfdv.

[0079] Step S25, traverse each operating state feature sf in the unit vector set trsdb of a single dry pump, and obtain the vacuum dry pump operating data distribution vector through the processing in step S24, which is in the form of [sfdv1, sfdv2, …, sfdv k ], where k=1,2,3,…,8.

[0080] Step S26, processing the downtime operation state vector set EDPVS in the vacuum dry pump operation state vector set DPVS in step S17 through steps S21 to S25 to obtain the downtime operation state data distribution vector set DPFI-E.

[0081] Step S27, processing the vacuum dry pump operation state vector set TDPVS to be tested in the vacuum dry pump operation state vector set DPVS in step S17 through steps S21 to S25 to obtain the vacuum dry pump operation state data distribution vector set DPFI-T to be predicted.

[0082] Step S28, marking the time corresponding to the first row of data of each processing unit in the downtime operation status data distribution vector set DPFI-E as the time of the unit distribution vector, and obtaining the downtime operation status label set EL.

[0083] Step S29, constructing the dry pump operation data distribution vector set DPFI={DPFI-E, EL, DPFI-T} through the obtained downtime operation status data distribution vector set DPFI-E, the vacuum dry pump operation status data distribution vector set DPFI-T to be predicted and the downtime operation status label set EL.

[0084] Step S3, constructing a vacuum dry pump distribution feature subset DPFS according to the vacuum dry pump operation state vector set DPVS and the feature subset-based data distribution vector set DPFI.

[0085] like Figure 4 As shown, step S3 specifically includes:

[0086] Step S31, obtaining the constructed second-level vector of the operating status of the vacuum dry pump, and selecting a subset svs of the second-level vector of the operating status of the vacuum dry pump.

[0087] Step S32, according to the vacuum dry pump operation status second-level vector subset svs, select the data distribution vector corresponding to the original feature from the downtime operation status data distribution vector set DPFI-E to obtain the downtime history data distribution feature subset DPFS-S.

[0088] Step S33, according to the second-level vector subset svs of the vacuum dry pump operation status, select the data distribution vector corresponding to the original feature from the data distribution vector set DPFI-T of the vacuum dry pump operation status to be predicted, and obtain the distribution feature subset DPFS-T of the vacuum dry pump to be predicted.

[0089] Step S34, obtain the downtime operation status label set EL, and then construct the dry pump distribution feature subset DPFS={DPFS-S, EL, DPFS-T} according to the downtime history data distribution feature subset DPFS-S, the predicted vacuum dry pump distribution feature subset DPFS-T and the downtime operation status label set EL.

[0090] Step S4, calculating the distances between the vacuum dry pump distribution feature subsets according to the vacuum dry pump distribution feature subset DPFS, and obtaining a distance list fs_dist of the vacuum dry pump distribution feature subsets to be predicted.

[0091] like Figure 5 As shown, step S4 specifically includes:

[0092] Step S41, obtaining a distribution feature subset DPFS-S of the downtime history data and a distribution feature subset DPFS-T of the vacuum dry pump to be predicted.

[0093] Step S42, selecting a feature subset vector corresponding to a certain device at a certain time in the feature subset DPFS-T of the vacuum dry pump distribution to be predicted.

[0094] Step S43, sequentially calculate the Euclidean distance of the feature subset vector at each moment in the downtime history data distribution feature subset DPFS-S, and obtain a list of vacuum dry pump distribution feature subset distances to be predicted fs_dist = [dist1, …, dist e ], where e=1,2,3,…,m.

[0095] Step S5, constructing a dry pump fault prediction training set FDDPD based on a distribution feature subset according to the data distribution vector set DPFI based on the feature subset and the distance list fs_dist of the vacuum dry pump distribution feature subset to be predicted.

[0096] like Figure 6As shown, step S5 specifically includes:

[0097] Step S51, sorting the obtained distance list fs_dist of the subset of distribution features of the vacuum dry pump to be predicted from small to large, and recording the sorted index index_fs.

[0098] Step S52, selecting the first k indexes in the sorted index index_fs, and constructing the first k most similar index sets MS_k.

[0099] Step S53, traverse all subsets in the predicted vacuum dry pump feature subset DPFS-T and execute steps S51 to S52 to obtain the most similar index set MS_k of the distribution feature subset v .

[0100] Step S54: The most similar index set MS_k is obtained. v Combine to get the most similar repeated index set Repeated-index v .

[0101] Step S55: The most similar repeated index set Repeated-index v Remove duplicate indexes to obtain the training set index train-index required for the dth downtime device at that moment d .

[0102] Step S56, taking out the downtime operation status data distribution vector set DPFI-E according to the training set instance index train-indexi in step S55, and obtaining a feature set FDDPD-data based on the distribution feature subset.

[0103] Step S57, taking out the downtime operation status label set EL according to the training set instance index train-indexi in step S55, and obtaining the label set FDDPD-label based on the distribution feature subset.

[0104] Step S58, constructing a vacuum dry pump fault prediction training set FDDPD={FDDPD-data, FDDPD-label} based on the distribution feature subset according to the feature set FDDPD-data based on the distribution feature subset and the label set FDDPD-label based on the distribution feature subset.

[0105] Step S6, using a Logistic classifier to establish a vacuum dry pump fault prediction model BFPM for the dry pump fault prediction training set FDDPD based on the distribution feature subset, and using the vacuum dry pump fault prediction model BFPM to perform vacuum dry pump fault prediction on other downtime pump data.

[0106] The most similar index set MS_k obtained under each distribution feature subset is set to 24*14, and the dry pump fault prediction training set FDDPD={FDDPD-data, FDDPD-label} based on the distribution feature subset is obtained under this selected number; the vacuum dry pump fault prediction model BFPM is established and predicted for the dry pump fault prediction training set FDDPD based on the distribution feature subset using the Logistic classifier. The experimental results show that the vacuum dry pump fault prediction model BFPM can use other downtime pump data to predict vacuum dry pump faults; thus demonstrating the effectiveness of the vacuum dry pump fault prediction method based on the distribution feature subset.

[0107] The data distribution vector of the vacuum dry pump at 17:00 on January 14, 2024 is used as the prediction vector. The number of historical distribution vectors selected under different k is as follows: Figure 7 As shown, where k=1,…,30.

[0108] The 62-day operation data of the vacuum dry pump is used as the model input, and the early warning situation obtained by using Logistic when k is set to 24*14 is as follows Figure 8 shown.

[0109] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some or all of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A vacuum dry pump fault prediction method based on a distribution feature subset, characterized in that: include: Step S1, constructing a vacuum dry pump operation state vector set DPVS; Step S2, constructing a data distribution vector set DPFI based on a feature subset according to the vacuum dry pump operation state vector set DPVS; Step S3, constructing a vacuum dry pump distribution feature subset DPFS according to the vacuum dry pump operation state vector set DPVS and the data distribution vector set DPFI based on the feature subset; Step S4, calculating the distances between the vacuum dry pump distribution feature subsets according to the vacuum dry pump distribution feature subset DPFS, and obtaining a distance list fs_dist of the vacuum dry pump distribution feature subsets to be predicted; Step S5, constructing a dry pump fault prediction training set FDDPD based on a distribution feature subset according to the data distribution vector set DPFI based on the feature subset and the distance list fs_dist of the vacuum dry pump distribution feature subset to be predicted; Step S6, using a Logistic classifier to establish a vacuum dry pump fault prediction model BFPM for the dry pump fault prediction training set FDDPD based on a distribution feature subset, and performing vacuum dry pump fault prediction on other downtime pump data through the vacuum dry pump fault prediction model BFPM; The step S1 comprises: Step S11, obtaining a set of operating data of vacuum dry pumps of the same type based on actual applications of vacuum dry pumps in the market; Step S12, constructing a vacuum dry pump operation data subset with different numbers under the same vacuum dry pump type; Step S13, constructing a second-level vector of the vacuum dry pump operation status {Time, DryPumpPower, BoosterPower, DPInverterSpeed, MBInverterSpeed, DPTemperature, MBTemperature, N2 Flow, ExhaustPressure} based on the dry pump motor power, motor speed, sensor temperature, nitrogen flow and exhaust pressure parameters; Step S14, the running status data set of the vacuum dry pump that has been shut down is processed by the step S13 to obtain a shut down running status vector set EDPVS = [rsv1, rsv2, …, rsv i , …, rsv n ], where i=1,2,3,…,n; Step S15, the operating state data set corresponding to all moments in the vacuum dry pump to be tested is processed by the step S13 to obtain the operating state vector set TDPVS of the vacuum dry pump to be tested = [rsvp1, rsvp2, …, rsvp j , …, rsvp m ], where j=1,2,3,…,m; Step S16, constructing the downtime operation status label set EL=[el1, el2, …,el x …, el n ], where x=1,2,3,…,n; the label set corresponds to each vacuum dry pump in the downtime operation state vector set ERSVS in step S14; Step S17, constructing a vacuum dry pump operation state vector set DPVS={EDPVS, EL, TDPVS} through the obtained downtime operation state vector set EDPVS, the vacuum dry pump operation state vector set TDPVS to be tested and the downtime operation state label set EL; In the step S13, Time represents the current time of the device operation, DryPumpPower represents the working power of the lower dry pump, BoosterPower represents the working power of the upper dry pump, DPInverterSpeed ​​represents the working frequency of the lower dry pump, MBInverterSpeed ​​represents the working frequency of the upper dry pump, DPTemperature represents the working temperature of the lower dry pump, MBTemperature represents the working temperature of the upper dry pump, N2 Flow represents the nitrogen flow rate introduced into the dry pump, and ExhaustPressure represents the exhaust pressure of the dry pump; The step S2 comprises: Step S21, setting the basic unit of data distribution vector processing to hours; Step S22, reading the operating status data trsd of a single vacuum dry pump; Step S23, the read operation status data trsd is processed by data distribution vector to obtain the unit vector set trsdb of a single dry pump, and the time corresponding to the first row of data in the unit vector set trsdb is used as the time of the trsdb, recorded as trsdb_t; Step S24, read a running state feature sf, calculate the minimum value, maximum value, average value, standard deviation and median of the running state feature sf, and use these five values ​​as the data distribution vector of the feature, recorded as sfdv; Step S25, traverse each operating state feature sf in the unit vector set trsdb of a single dry pump, and obtain the vacuum dry pump operating data distribution vector through the processing in step S24, which is in the form of [sfdv1, sfdv2, …, sfdv k …, sfdv8], where k = 1, 2, 3, …, 8; Step S26, processing the downtime operation state vector set EDPVS in the vacuum dry pump operation state vector set DPVS in the step S17 through the steps S21 to S25 to obtain the downtime operation state data distribution vector set DPFI-E; Step S27, processing the vacuum dry pump operation state vector set TDPVS to be tested in the vacuum dry pump operation state vector set DPVS in the step S17 through the steps S21 to S25 to obtain a vacuum dry pump operation state data distribution vector set DPFI-T to be predicted; Step S28, marking the time corresponding to the first row of data of each processing unit in the downtime operation status data distribution vector set DPFI-E as the time of the unit distribution vector, and obtaining the downtime operation status label set EL; Step S29, constructing the dry pump operation data distribution vector set DPFI={DPFI-E, EL, DPFI-T} through the obtained downtime operation status data distribution vector set DPFI-E, the vacuum dry pump operation status data distribution vector set DPFI-T to be predicted and the downtime operation status label set EL.

2. A vacuum dry pump fault prediction method based on distribution feature subset according to claim 1, characterized in that: The step S3 comprises: Step S31, obtaining the constructed second-level vector of the operating state of the vacuum dry pump, and selecting a subset svs in the second-level vector of the operating state of the vacuum dry pump; Step S32, selecting the data distribution vector corresponding to the original feature from the downtime operation status data distribution vector set DPFI-E according to the vacuum dry pump operation status second-level vector subset svs, and obtaining the downtime history data distribution feature subset DPFS-S; Step S33, selecting the data distribution vector corresponding to the original feature from the data distribution vector set DPFI-T of the vacuum dry pump operating state to be predicted according to the second-level vector subset svs of the vacuum dry pump operating state, and obtaining the distribution feature subset DPFS-T of the vacuum dry pump to be predicted; Step S34, obtain the downtime operation status label set EL, and then construct the dry pump distribution feature subset DPFS={DPFS-S, EL, DPFS-T} according to the downtime history data distribution feature subset DPFS-S, the predicted vacuum dry pump distribution feature subset DPFS-T and the downtime operation status label set EL.

3. A vacuum dry pump fault prediction method based on distribution feature subset according to claim 2, characterized in that: The step S4 comprises: Step S41, obtaining a distribution feature subset DPFS-S of downtime history data and a distribution feature subset DPFS-T of a vacuum dry pump to be predicted; Step S42, selecting a feature subset vector corresponding to a certain device at a certain time in the vacuum dry pump distribution feature subset DPFS-T to be predicted; Step S43, sequentially calculate the Euclidean distance of the feature subset vector at each moment in the downtime history data distribution feature subset DPFS-S, and obtain a list of vacuum dry pump distribution feature subset distances to be predicted fs_dist = [dist1, ..., dist e …, dist m ], where e=1,2,3,…,m.

4. A vacuum dry pump fault prediction method based on distribution feature subset according to claim 3, characterized in that: The step S5 comprises: Step S51, sorting the obtained distance list fs_dist of the distribution feature subset of the vacuum dry pump to be predicted from small to large, and recording the sorted index index_fs; Step S52, selecting the first k indexes in the sorted index index_fs, and constructing the first k most similar index set MS_k; Step S53, traverse all subsets in the vacuum dry pump feature subset DPFS-T to be predicted and execute the steps S51 to S52 to obtain the most similar index set MS_kv of the distribution feature subset; Step S54, combining the obtained most similar index set MS_kv to obtain the most similar repeated index set Repeated-indexv; Step S55, removing repeated indexes from the most similar repeated index set Repeated-indexv to obtain the training set index train-indexd required in the dth downtime device at this moment; Step S56, taking out the downtime operation status data distribution vector set DPFI-E according to the training set instance index train-indexi of step S55, and obtaining a feature set FDDPD-data based on the distribution feature subset; Step S57, taking out the downtime operation status label set EL according to the training set instance index train-indexi of step S55, and obtaining the label set FDDPD-label based on the distribution feature subset; Step S58, constructing a vacuum dry pump fault prediction training set FDDPD={FDDPD-data, FDDPD-label} based on the distribution feature subset according to the feature set FDDPD-data based on the distribution feature subset and the label set FDDPD-label based on the distribution feature subset.

Images