CN103323103A - Real-time prediction method for low-frequency vibration of large steam turbine generator unit - Google Patents

Abstract

The invention discloses a real-time prediction method for low-frequency vibration of a large steam turbine generator unit, and belongs to the field of rotary mechanical vibration condition monitoring and fault diagnosis. The method comprises the steps that low-frequency vibration data in axial relative vibration of a rotor of the steam turbine generator unit are collected, calculated and stored, on this basis, an FFT spectral analysis method and an ANFIS self-adaption nerve fuzzy inference system computing method are combined, and the amplitude of the low-frequency vibration is predicted and calculated in real time. According to the real-time prediction method for the low-frequency vibration of the large steam turbine generator unit, the vibration data of the rotor when the generator unit is in operation and the ANFIS method are utilized, by being calculated and predicted, the amplitude data of the low-frequency vibration of the rotor can be monitored, analyzed and distinguished in an automatic, real-time and online mode, and the efficiency and the accuracy of the work of monitoring, predicting and analyzing the low-frequency vibration of the large steam turbine generator unit in real time are improved.

Description

The Low Frequency Vibration in Large Turbine Generator Sets real-time predicting method

Technical field

The invention belongs to rotating machinery vibrating condition monitoring and fault diagnosis field, relate in particular to a kind of Low Frequency Vibration in Large Turbine Generator Sets real-time predicting method.

Background technology

Large turbo-type generator group axle is that low-frequency vibrating failure is a kind of asynchronous vibration, shows as faults coupling strong, and the failure mechanism that relates to is complicated.Large turbo-type generator group axle is the low-frequency vibration coupling fault owing to having Non-linear coupling between the various faults, and therefore more complicated than the single failure rotor, rotor-support-foundation system can show more complicated, abundanter non-linear phenomena.

If can make certain prediction to generation and the development of unit low-frequency vibrating failure, just take necessary maintenance measure in commitment or budding stage that low-frequency vibrating failure occurs, realize initial failure identification, indication, rather than after fault worsens, go again to solve to process, the prevention low-frequency vibrating failure instructs according to being very significant in " possible trouble " and for unit maintenance provides.

The low-frequency vibration of discriminatory analysis machine group rotor is finished by the professional with certain field operation experiences and professional knowledge technical ability usually, can't accomplish the astable real-time automatic on-line monitoring of low-frequency vibration, analyzes and differentiate.Therefore, propose a kind of Low Frequency Vibration in Large Turbine Generator Sets method for quick predicting and just seem very important.

Summary of the invention

During for the discriminatory analysis machine group rotor low-frequency vibration mentioned in the background technology, can't accomplish the problems such as the astable real-time automatic on-line monitoring of low-frequency vibration, analysis and differentiation, the present invention proposes a kind of Low Frequency Vibration in Large Turbine Generator Sets real-time predicting method.

A kind of Low Frequency Vibration in Large Turbine Generator Sets real-time predicting method is characterized in that, described method specifically may further comprise the steps:

Step 1: adopt the axle Relative Vibration data of vibration at high speed data collecting card Real-time Collection rotor of turbogenerator set one side radial journal bearing, tach signal and the key signal of rotor;

Step 2: utilize the FFT frequency spectrum analysis method that the axle Relative Vibration data of current time radial journal bearing are carried out spectrum analysis, obtain the corresponding vibration amplitude data sequence of different vibration frequencies from the low frequency to the high frequency;

Step 3: from the vibration amplitude data sequence that step 2 obtains, all Frequencies less than unit working speed respective frequencies of intercepting current time, and calculate vibration amplitude sum A corresponding to all Frequencies; Storage low-frequency vibration amplitude A is every Δ _sStorage second once;

Step 4: judge that whether low-frequency vibration amplitude storage time is greater than Preset Time segment length P _MNIf greater than the Preset Time segment length, record so current time T _NFront low-frequency vibration amplitude data A carries out subsequent calculations and judges; Otherwise, continue the storage data;

Step 5: according to the ordering of time data memory sequencing, different constantly lower low-frequency vibration amplitude data represent the time data memory sequencing with subscript i, i=1, and 2,3 ..., m; From T _NConstantly be truncated to forward T _MLow-frequency vibration amplitude data A constantly is with T _MConstantly to T _NLow-frequency vibration amplitude data constantly are expressed as A _i(i=1,2,3 ..., m); | T _N-T _M|=p _MN, p _MNBe the Preset Time segment length;

Step 6: according to low-frequency vibration amplitude data A _i(i=1,2,3 ..., m), set up ANFIS(ANFIS (adaptive neuro-fuzzy inference system, Adaptive Neuro-fuzzy Inference) prediction learning training ordered series of numbers and train, obtain the rule group R of fuzzy inference system FIS;

Step 7: according to any current time low frequency amplitude data

And front low-frequency vibration amplitude data, specifically comprise Utilize rule group R, carry out calculating based on the fuzzy reasoning of ANFIS method, prediction and calculation obtains p * Δ _sLow-frequency vibration amplitude data after second

Wherein, j is the item number of ANFIS prediction and calculation data input item, and p is prediction step; s _jSubscript for low-frequency vibration amplitude data; s _j+ p is p * Δ _sLow-frequency vibration amplitude data after second

Subscript.

Described unit working speed respective frequencies is 50 hertz.

Described according to low-frequency vibration amplitude data A _i(i=1,2,3 ..., m), set up the process that ANFIS prediction learning training ordered series of numbers trains and be:

Step a: set the item number j of ANFIS prediction and calculation data input item, prediction step is p; The subscript sequence of first trip is [s ₁s ₂S _J-1s _js _j+ p]; Total line number of ANFIS prediction learning training ordered series of numbers is r, r=m-s _j-p+1; ANFIS prediction learning training ordered series of numbers is:

$\left[\begin{array}{cccccc}{A}_{s_1}& A_{s_2}& ...& A_{s_{j-1}}& A_{s_j}& A_{s_j+p}\\ A_{s_1+1}& A_{s_2+1}& ...& A_{s_{j-1}+1}& A_{s_j+1}& A_{s_j+p+1}\\ ...& ...& ...& ...& ...& ...\\ A_{s_1+r-2}& A_{s_2+r-2}& ...& A_{s_{j-1}+r-2}& A_{s_j+r-2}& A_{m-1}\\ A_{s_1+r-1}& A_{s_2+r-1}& ...& A_{s_{j-2}+r-1}& A_{s_j+r-1}& A_m\end{array}\right]$

Wherein, last row of every row are training study output items, and all the other are training study input items;

Step b: the training study calculating parameter of setting ANFIS comprises: the number of setting frequency of training and each input item subordinate function and subordinate function; Utilize ANFIS prediction learning training ordered series of numbers to carry out Adaptive Neuro-fuzzy Inference ANFIS training.

The invention has the beneficial effects as follows, the Low Frequency Vibration in Large Turbine Generator Sets real-time predicting method that proposes, utilize unit operation rotor vibration data and ANFIS method, through calculating prediction rotor low-frequency vibration amplitude data, can realize automatic real time on-line monitoring, analysis and distinguishing, improve efficient and the accuracy of Low Frequency Vibration in Large Turbine Generator Sets Real-Time Monitoring forecast analysis work.

Description of drawings

Fig. 1 is Low Frequency Vibration in Large Turbine Generator Sets real-time predicting method functional flow diagram provided by the invention;

Fig. 2 is Turbo-generator Set low-frequency vibration real-time estimate schematic diagram provided by the invention.

Embodiment

Below in conjunction with accompanying drawing, preferred embodiment is elaborated.Should be emphasized that, following explanation only is exemplary, rather than in order to limit the scope of the invention and to use.

A kind of Low Frequency Vibration in Large Turbine Generator Sets real-time predicting method is characterized in that, described method specifically may further comprise the steps as shown in Figure 1:

Step 1: real time data acquisition calculates storage

Adopt near the axle Relative Vibration data that record the vibration at high speed data collecting card Real-time Collection rotor of turbogenerator set one side radial journal bearing, tach signal and the key signal of rotor.Each passage technology parameter of vibrating data collection card is 50ks/s, 24bit.

Step 2: for the axle Relative Vibration data of unit high pressure rotor one side, utilize the FFT frequency spectrum analysis method, calculate the corresponding vibration amplitude data sequence of the different vibration frequencies of current time from the low frequency to the high frequency (amplitude unit is μ m).Therefrom intercepting obtains frequency less than all low-frequency vibration amplitudes of current time of unit working speed respective frequencies (50Hz), and calculates all low-frequency vibration amplitude sum A;

Step 3: storage low-frequency vibration amplitude A, every storage in 1 second once;

Step 4: whether judge low-frequency vibration amplitude storage time greater than 500 seconds, if the time of storage data greater than 500 seconds, is recorded current time T so _NFront low-frequency vibration amplitude data A carries out subsequent calculations and judges; Otherwise, continue the storage data;

Step 5: according to the current time T that has stored _NFront low-frequency vibration amplitude data A is from T _NConstantly be truncated to forward T _MLow-frequency vibration amplitude data A constantly, | T _N-T _M|=p _MN, p _MNBe Preset Time segment length, p _MN=500 seconds.According to the ordering of time data memory sequencing, different constantly lower low-frequency vibration amplitude data represent the time data memory sequencing with subscript i.Therefore, T _MConstantly to T _NLow-frequency vibration amplitude data constantly can be expressed as A _i(i=1,2,3 ..., 500).

According to low-frequency vibration amplitude data A _i(i=1,2,3 ..., 500), set up ANFIS prediction learning training ordered series of numbers.At first, determine the item number 6 of ANFIS prediction and calculation data input item, prediction step is respectively 1,2,3.

The subscript sequence of determining first trip is respectively:

[1?2?3?4?5?6?7]，

[1?2?3?4?5?6?8]，

[1?2?3?4?5?6?9]。

Total line number of ANFIS prediction learning training ordered series of numbers is 494,493,492.

According to T _MConstantly to T _NLow-frequency vibration amplitude data A constantly _i(i=1,2,3 ..., 500), form 3 ANFIS prediction learning training ordered series of numbers as follows:

$\left[\begin{array}{ccccccc}{A}_1& A_2& A_3& A_4& A_5& A_6& A_7\\ A_2& A_3& A_4& A_5& A_6& A_7& A_8\\ ...& ...& ...& ...& ...& ...& ...\\ A_{493}& A_{494}& A_{495}& A_{496}& A_{497}& A_{498}& A_{499}\\ A_{494}& A_{495}& A_{496}& A_{497}& A_{498}& A_{499}& A_{500}\end{array}\right],$

$\left[\begin{array}{ccccccc}{A}_1& A_2& A_3& A_4& A_5& A_6& A_8\\ A_2& A_3& A_4& A_5& A_6& A_7& A_9\\ ...& ...& ...& ...& ...& ...& ...\\ A_{492}& A_{493}& A_{494}& A_{495}& A_{496}& A_{497}& A_{499}\\ A_{493}& A_{494}& A_{495}& A_{496}& A_{497}& A_{498}& A_{500}\end{array}\right],$

$\left[\begin{array}{ccccccc}{A}_1& A_2& A_3& A_4& A_5& A_6& A_9\\ A_2& A_3& A_4& A_5& A_6& A_7& A_{10}\\ ...& ...& ...& ...& ...& ...& ...\\ A_{491}& A_{492}& A_{493}& A_{494}& A_{495}& A_{496}& A_{499}\\ A_{492}& A_{493}& A_{494}& A_{495}& A_{496}& A_{497}& A_{500}\end{array}\right].$

In the above-mentioned ordered series of numbers, front 6 row of every row are training study input items, and last row are training study output items.Setting frequency of training is 10, gbell type subordinate function, for each input item is set 2 subordinate functions, based on above-mentioned basic setting, utilize above-mentioned 3 ANFIS prediction learning training ordered series of numbers, carry out ANFIS (adaptive neuro-fuzzy inference system, Adaptive Neuro-fuzzy Inference) learning training and calculate, obtain FIS (fuzzy inference system, fuzzy inference system) rule group R ¹, R ², R ³, the regular number of each rule group is 2 ⁶Individual.(the ANFIS computing method are general mathematic calculation that the professional knows).

Step 6: according to any current time K and front low-frequency vibration amplitude data thereof, specifically comprise Utilize respectively rule group R ¹, R ², R ³, carry out calculating based on the fuzzy reasoning of ANFIS method, can predict the low-frequency vibration amplitude data that obtain after 1 second, 2 seconds, 3 seconds (the fuzzy reasoning calculating based on the ANFIS method is the general mathematic calculation that the professional knows).

According to current time and front low-frequency vibration amplitude data (amplitude unit is μ m) thereof, specifically comprise 21.8,22.4,17.7,22.4,20.7,17.5, utilize respectively rule group R ¹, R ², R ³, carry out calculating based on the fuzzy reasoning of ANFIS method, can predict the low-frequency vibration amplitude data 18.6,20.2,22.7 that obtain after 1 second, 2 seconds, 3 seconds.

Fig. 2 is Turbo-generator Set low-frequency vibration real-time estimate schematic diagram provided by the invention.Among Fig. 2, the tach signal of armature spindle Relative Vibration data, rotor and key signal can obtain from the supervisory instrument (TSI) of configuration Turbo-generator Set, and power of the assembling unit data-signal can obtain from the dcs (DCS) of configuration Turbo-generator Set.In the present embodiment, the tach signal of armature spindle Relative Vibration data, rotor and key signal are supervisory instrument (TSI) acquisitions from the configuration Turbo-generator Set.Turbo-generator Set low-frequency vibration real-time estimate schematic diagram is illustrated in fig. 2 shown below, in the slot that data collecting card insertion industrial microcomputer (IPC) provides.Requirement according to data collecting card, the data acquisition conditioning device is processed axle Relative Vibration signal, the tach signal of rotor, the key signal from Turbo-generator Set supervisory instrument (TSI), the vibration at high speed data collecting card in axle Relative Vibration signal after treatment, the tach signal of rotor, the key signal input IPC.Each passage technology parameter of vibrating data collection card is 50ks/s, 24bit.

The above; only for the better embodiment of the present invention, but protection scope of the present invention is not limited to this, anyly is familiar with those skilled in the art in the technical scope that the present invention discloses; the variation that can expect easily or replacement all should be encompassed within protection scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with the protection domain of claim.

Claims (3)

1. a Low Frequency Vibration in Large Turbine Generator Sets real-time predicting method is characterized in that, described method specifically may further comprise the steps:

Step 1: adopt the axle Relative Vibration data of vibration at high speed data collecting card Real-time Collection rotor of turbogenerator set one side radial journal bearing, tach signal and the key signal of rotor;

Step 2: utilize the FFT frequency spectrum analysis method that the axle Relative Vibration data of current time radial journal bearing are carried out spectrum analysis, obtain the corresponding vibration amplitude data sequence of different vibration frequencies from the low frequency to the high frequency;

Step 3: from the vibration amplitude data sequence that step 2 obtains, all Frequencies less than unit working speed respective frequencies of intercepting current time, and calculate vibration amplitude sum A corresponding to all Frequencies; Storage low-frequency vibration amplitude A is every Δ _sStorage second once;

Step 4: judge that whether low-frequency vibration amplitude storage time is greater than Preset Time segment length P _MNIf greater than the Preset Time segment length, record so current time T _NFront low-frequency vibration amplitude data A enters step 5; Otherwise, continue the storage data;

Step 5: according to the ordering of time data memory sequencing, different constantly lower low-frequency vibration amplitude data represent the time data memory sequencing with subscript i, i=1, and 2,3 ..., m; From T _NConstantly be truncated to forward T _MLow-frequency vibration amplitude data A constantly is with T _MConstantly to T _NLow-frequency vibration amplitude data constantly are expressed as A _i(i=1,2,3 ..., m); | T _N-T _M|=p _MN, p _MNBe the Preset Time segment length;

Step 6: according to low-frequency vibration amplitude data A _i(i=1,2,3 ..., m), set up ANFIS prediction learning training ordered series of numbers and train, obtain the rule group R of fuzzy inference system FIS;

Step 7: according to any current time low frequency amplitude data

And front low-frequency vibration amplitude data, specifically comprise

Utilize rule group R, carry out calculating based on the fuzzy reasoning of ANFIS method, prediction and calculation obtains p * Δ _sLow-frequency vibration amplitude data after second

Wherein, j is the item number of ANFIS prediction and calculation data input item, and p is prediction step; s _jSubscript for low-frequency vibration amplitude data; s _j+ p is p * Δ _sLow-frequency vibration amplitude data after second

Subscript.

2. method according to claim 1 is characterized in that, described unit working speed respective frequencies is 50 hertz.

3. method according to claim 1 is characterized in that, and is described according to low-frequency vibration amplitude data A _i(i=1,2,3 ..., m), set up the process that ANFIS prediction learning training ordered series of numbers trains and be:

Step a: set the item number j of ANFIS prediction and calculation data input item, prediction step is p; The subscript sequence of first trip is [s ₁s ₂S _J-1s _js _j+ p]; Total line number of ANFIS prediction learning training ordered series of numbers is r, r=m-s _j-p+1; ANFIS prediction learning training ordered series of numbers is:

$\left[\begin{array}{cccccc}{A}_{s_1}& A_{s_2}& ...& A_{s_{j-1}}& A_{s_j}& A_{s_j+p}\\ A_{s_1+1}& A_{s_2+1}& ...& A_{s_{j-1}+1}& A_{s_j+1}& A_{s_j+p+1}\\ ...& ...& ...& ...& ...& ...\\ A_{s_1+r-2}& A_{s_2+r-2}& ...& A_{s_{j-1}+r-2}& A_{s_j+r-2}& A_{m-1}\\ A_{s_1+r-1}& A_{s_2+r-1}& ...& A_{s_{j-2}+r-1}& A_{s_j+r-1}& A_m\end{array}\right]$

Wherein, last row of every row are training study output items, and all the other are training study input items;

Step b: the training study calculating parameter of setting ANFIS comprises: the number of setting frequency of training and each input item subordinate function and subordinate function; Utilize ANFIS prediction learning training ordered series of numbers to carry out Adaptive Neuro-fuzzy Inference ANFIS training.

Images