(19)
(11)EP 3 196 657 B1

(12)EUROPEAN PATENT SPECIFICATION

(45)Mention of the grant of the patent:
11.12.2019 Bulletin 2019/50

(21)Application number: 17151942.4

(22)Date of filing:  18.01.2017
(51)International Patent Classification (IPC): 
G01P 3/48(2006.01)
G01M 15/12(2006.01)
G01P 3/44(2006.01)

(54)

SPEED ESTIMATION SYSTEMS

GESCHWINDIGKEITSSCHÄTZUNGSSYSTEME

SYSTÈMES D'ESTIMATION DE VITESSE


(84)Designated Contracting States:
AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

(30)Priority: 20.01.2016 US 201615002364

(43)Date of publication of application:
26.07.2017 Bulletin 2017/30

(73)Proprietor: Simmonds Precision Products, Inc.
Vergennes, VT 05491 (US)

(72)Inventors:
  • LIU, Lei
    Shelburne, VT 05482 (US)
  • BRUCE, Kyle M.
    Burlington, VT 05401 (US)

(74)Representative: Dehns 
St. Bride's House 10 Salisbury Square
London EC4Y 8JD
London EC4Y 8JD (GB)


(56)References cited: : 
US-A- 4 310 892
US-A1- 2014 281 779
US-A1- 2005 209 814
  
  • YUANXI YANG ET AL: "An Adaptive Kalman Filter Based on Sage Windowing Weights and Variance Components", JOURNAL OF NAVIGATION., vol. 56, no. 2, 1 May 2003 (2003-05-01), pages 231-240, XP055371010, GB ISSN: 0373-4633, DOI: 10.1017/S0373463303002248
  
Note: Within nine months from the publication of the mention of the grant of the European patent, any person may give notice to the European Patent Office of opposition to the European patent granted. Notice of opposition shall be filed in a written reasoned statement. It shall not be deemed to have been filed until the opposition fee has been paid. (Art. 99(1) European Patent Convention).


Description

BACKGROUND


1. Field



[0001] The present disclosure relates to vibration signal processing systems and methods, more specifically to rotation speed and/or imbalance tracking for vibratory systems.

2. Description of Related Art



[0002] Estimating rotation speed through vibration is an important step in vibration based mechanical diagnostics and prognostics applications. Reliable speed estimation is rudimental to vibration data analysis and algorithmic processing since many failure signatures are highly correlated to the rotation speed, or the fundamental frequency. As the underlying physics of vibration analysis finds the periodic component existing in vibration associated with imbalanced rotation, correct speed estimations will also allow vibration due to imbalance be appropriately pinpointed, measured, and thus minimized with appropriate countermeasures without the need of a tachometer.

[0003] While many approaches for estimating fundamental frequency have been developed in various closely related fields, none have been able to give correct results once signal-to-noise ratio (SNR) drops to a certain degree. As implied by the ratio, a drop of SNR includes two potential factors: the increase of noises and the decrease of signals. Conventional or adaptive filters are often used to address the former when noises have known characteristics or are able to be referenced. However, non-stationary noises (e.g., those that emerge and fade out due to sudden changes in external environment) cannot be effectively attenuated. Worsened by the later factor, speed estimations can also become erroneous due to changes of operational conditions internally within systems (e.g., ramping up/down and loading shift). Spurious speed estimations that occur in these transient changes need to be identified and excluded before imbalance and conditions are algorithmically assessed.

[0004] Prior art sensing systems in which vibration is detected are known from US 4310892 and US 2005/0209814, and the use of Kalman filters is described in US 4310892, US 2014/0281779 and the article "An Adaptive Kalman Filter Based on Sage Windowing Weights and Variance Components" by Yuanxi Yang et al in "The Journal of Navigation", vol. 56, no. 2, pages 231 - 240, May 2003.

[0005] Such conventional methods and systems have generally been considered satisfactory for their intended purpose. However, there is still a need in the art for improved speed estimation systems and methods. The present disclosure provides a solution for this need.

SUMMARY



[0006] According to a first aspect of the present invention, there is provided a method for estimating rotational speed of a system, comprising: receiving vibrational data from a sensor; estimating a speed from the vibrational data to create estimated speed data; and filtering the estimated speed data through an adaptively weighted filter to minimize incorrect speed estimation; wherein filtering the estimated speed data through the adaptively weighted filter includes filtering the estimated speed data through a Kalman filter which includes a Kalman gain Kk, wherein

wherein Pk- is an a priori prediction error covariance, R is a covariance matrix of measurement noises, and w is the adaptive weight; wherein filtering the estimated speed data through the Kalman filter includes using a test statistic to measure the deviation of current estimation from previous estimations to enforce continuity in time domain, including determining errors between speed estimation and at least one of a speed prediction, a standard score, or a Mahalanobis distance to determine an estimation correctness and to modify the adaptive weight w if incorrect estimation is determined.

[0007] Modifying the adaptive weight w can include modifying the adaptive weight w in real time. Modifying the adaptive weight can include comparing a Mahalanobis distance D to a threshold T, and setting a value for the adaptive weight w such that,





[0008] According to a further aspect of the present invention, there is provided a speed estimation system for a rotational system, comprising: a vibrational sensor configured to output vibrational signals; a speed estimation module operatively connected to the vibrational sensor to receive the vibrational signals and output estimated speed data based on the vibrational signals; and an adaptively weighted filter module configured to receive the estimated speed data and configured to output filtered estimated speed data, wherein the adaptively weighted filter module includes a Kalman filter having an adaptive weight, the Kalman filter including a Kalman gain Kk, wherein

wherein Pk- is an a priori prediction error covariance, R is a covariance matrix of measurement noises, and w is the adaptive weight; wherein filtering the estimated speed data through the Kalman filter includes using a test statistic to measure the deviation of current estimation from previous estimations to enforce continuity in time domain, including determining errors between speed estimation and at least one of a speed prediction, a standard score, or a Mahalanobis distance to determine an estimation correctness and to modify the adaptive weight w if incorrect estimation is determined.

[0009] In certain embodiments, the system can form part of or can be operatively connected to a controller which controls one or more inputs to the rotational system to provide feedback to the controller.

[0010] These and other features of the systems and methods of the subject disclosure will become more readily apparent to those skilled in the art from the following detailed description taken in conjunction with the drawings.

BRIEF DESCRIPTION OF THE DRAWINGS



[0011] So that those skilled in the art to which the subject disclosure appertains will readily understand how to make and use the devices and methods of the subject disclosure without undue experimentation, embodiments thereof will be described in detail herein below with reference to certain figures, wherein:

Fig. 1 is a schematic diagram of an embodiment of a system in accordance with this disclosure;

Fig. 2A is a chart showing speed estimated from a vibration (e.g., a single channel) when a rotational system under observation is in volatile operation;

Fig. 2B is a chart showing a Kalman filter applied to the estimated speeds of Fig. 2A to smooth results; and

Fig. 2C is a chart showing speed estimation with rejection after applying an adaptively weighted Kalman filter to the estimated speed in Fig. 2A in accordance with this disclosure.


DETAILED DESCRIPTION



[0012] Reference will now be made to the drawings wherein like reference numerals identify similar structural features or aspects of the subject disclosure. For purposes of explanation and illustration, and not limitation, an illustrative view of an embodiment of a system in accordance with the disclosure is shown in Fig. 1 and is designated generally by reference character 100. Other embodiments and/or aspects of this disclosure are shown in Figs. 2A to 2C. The systems and methods described herein can be used to more accurately estimate speed.

[0013] Referring to Fig. 1, a speed estimation system 100 for a rotational system (not shown) can include at least one vibrational sensor 101 configured to output vibrational signals. The vibrational sensor 101 can include any suitable sensor (e.g., an accelerometer). The at least one sensor 101 can be mounted in any suitable location (e.g., externally mounted to a vibrational system such as a housing of a compressor).

[0014] The system 100 can include a speed estimation module 103 operatively connected to the vibrational sensor 101 to receive the vibrational signals and output estimated speed data based on the vibrational signals. The speed estimation 103 can estimate speed based on the vibrations of the rotational system via any suitable speed estimation methods/models (e.g., as described herein below).

[0015] The system 100 also includes an adaptively weighted filter module 105 configured to receive the estimated speed data and configured to output filtered estimated speed data. The adaptively weighted filter module 105 can filter the estimated speed date using any suitable methods/models (e.g., a weighted Kalman filter as described hereinbelow). In certain embodiments, the system 100 can form part of or can be operatively connected to a controller 99 which controls one or more inputs to the rotational system to provide feedback to the controller 99.

[0016] Using the fact that rotation speed can only change continually in a mechanical system, spurious speed estimations can be rejected by constantly tracking the estimations and using speed continuity as a criteria to determine the validity of individual estimation. This imposes a layer of continuity enforcement in time domain on top of estimation results, and it can be viewed as a low pass filtering gate with cut-off frequency being able to be adapted to the physics of systems of interests. To this end, a Kalman filter can be routinely used to provide predictions to be compared with estimations to obtain reliable results.

[0017] Unfortunately, predictions from a traditional Kalman filter are subject to heavy influences from outliers, namely those previously incorrect speed estimations. This is because the Kalman filter is linear and built upon normality assumption, therefore a single outlier will spoil many of the subsequent predictions. A direct consequence in speed estimation is that many correct estimation results following an incorrect estimation can be falsely identified as incorrect. This is worsened when transient noises/changes are highly frequent, and may render no output from the filter at all.

[0018] A discrete time model for tracking changes of speed estimation can be written as

where vk is a vector of estimated speeds at time instant k, vk-1 is a vector of speeds at time instant k-1, u is the speed changes between the time instants, and w is the inaccuracies or noises due to the estimation of speeds from vibrations. For a single vibration channel, the model simply reduces to a univariate system.

[0019] The model can be easily rearranged into the standard Kalman filter model by setting both the state transition matrix and the observation matrix to be identity matrices and further treating the speed change u as a process noise. It is apparent that the two noise terms are independent of each other, thereafter they are assumed to be normally distributed as p(u)∼N(0, Q) and p(w)N(0,R), respectively.

[0020] The model has a strong physical basis. On one hand, the variance of the process noise, Q, is directly related to the specification of underlying mechanical systems and can be easily quantified. That is, under normal operation conditions, the maximum allowable change of speed is usually bounded by specification and therefore can be confidently used for Q. On the other hand, the covariance matrix R of the measurement noises is reduced to a diagonal matrix following the assumption on inter-channel independence. The variance of each vibration channel, or the diagonal element of R, may be further quantified by the resolution of the speed estimation algorithm, which is often theoretically attainable. In a typical application that all vibration channels use the same algorithm, R may have identical diagonal elements.

[0021] Using a Kalman filter, a prediction of speed can be obtained by the current speed estimation and the prediction from the previous step. The recursive step can be expressed as,

where ṽk is the current predicted speed value, ṽk-1 is the preceding predicted speed value, vk is the estimated speed value, and Kk is the Kalman gain. Measuring the distance between the prediction and the estimation provides a way to reject spurious estimations. However, as previously stated, a single outlier in estimation will affect several future predictions. This can be justified by examining the Kalman gain simplified for model (1) and written in a set of recursive equations as,

where Pk- is the current a priori prediction error covariance, Pk-1- is the preceding a priori prediction error covariance, Pk-1 is the preceding a posteriori prediction error covariance, Kk-1 is the preceding Kalman gain, R is the covariance matrix of the measurement noises, and I is the identity matrix. It can be seen from these equations that, if Q and R remain constant per the model, both Kalman gain K and prediction error covariance P stabilize over time and converge to constants as well. An outlier in estimation will invalidate the implication of Q in the model such that its residual effect only slowly disappears in proceeding predictions.

[0022] To improve robustness and accuracy, the Kalman gain is modified by introducing an adaptive weight, w, into Eq. (3) as,



[0023] This can be understood by noticing that the noise covariance is equivalently increased in response to an outlier. Consequently, K and P will not converge to constants anymore but be adaptive. It would be expected that a suitable adaptive weight would make the Kalman gain less sensitive to a predefined noise covariance R when an outlier is detected, meanwhile this can allow the Kalman gain to restore to its optimum value quickly if the outliers are no longer present.

[0024] To detect outliers and to define an adaptive weight, certain test statistics can be utilized. For example the Mahalanobis distance for multiple vibration channels can be utilized. The test statistic quantifies the errors between speed estimation and prediction, and therefore can be compared to a threshold T such that,



[0025] In the case of a single vibration channel, the Mahalanobis distance D reduces to a standard score noticing that Pk- is in fact the prediction error variance. The selection of the threshold T becomes trivial in this case as 3 is a natural choice.

[0026] Referring to Figs. 2A, 2B, and 2C, an embodiment of speed estimation data is shown without filtering (in Fig. 2A), with a traditional Kalman filter (in Fig. 2B), and with an embodiment of an adaptively weighted Kalman filter as described above (in Fig. 2C). As can be seen, the results of the adaptively weighted Kalman filter track a much more accurate and realistic speed estimation.

[0027] Any suitable portion or the entirety of the above described methods and systems can be implemented via any suitable computer hardware (e.g., a microprocessor, a memory) or software (of any suitable language). In certain embodiments, any suitable portion or the entirety of the above described methods and systems can form a part of or be operative with a controller connected to the vibratory system (e.g., a compressor), such that the controller can control one or more inputs (e.g., speed) to the vibratory system.

[0028] As described above, a speed estimating system having a robust adaptively weighted filter (e.g., the above described weighted Kalman filter) can be designed to minimize the adverse consequence from incorrect speed estimations. A test statistic of errors between speed estimation and prediction, standard score or Mahalanobis distance, can be used to decide the estimation correctness and further to modify the filter (e.g., Kalman filter gain) if incorrect estimation is determined. The test statistic can measure the deviation of current estimation from previous noisy estimations, therefore providing a way to enforce continuity in time domain. In the meantime, the modified gain makes predictions less sensitive to preceding incorrect estimations. The accurate speed estimation results can then be used for determining if there is imbalanced vibration and other condition indicators in real time (e.g., for a compressor, fan, or any other suitable device).

[0029] Restated, rotation speeds can be continuously tracked in real time, and spurious speed estimations can be rejected in an early stage so as to not contaminate speed measurements and associated operational conditions, e.g. imbalanced vibration. Embodiments can reduce artifacts/noises in failure signatures that are closely related to those measurements when a speedometer or tachometer is not available.

[0030] The methods and systems of the present disclosure, as described above and shown in the drawings, provide for speed estimation systems with superior properties including improved speed estimation.


Claims

1. A method for estimating rotational speed of a system, comprising:

receiving vibrational data from a sensor;

estimating a speed from the vibrational data to create estimated speed data; and

filtering the estimated speed data through an adaptively weighted filter to minimize incorrect speed estimation;

wherein filtering the estimated speed data through the adaptively weighted filter includes filtering the estimated speed data through a Kalman filter which includes a Kalman gain Kk, wherein

wherein Pk- is an a priori prediction error covariance, R is a covariance matrix of measurement noises, and w is the adaptive weight;

wherein filtering the estimated speed data through the Kalman filter includes using a test statistic to measure the deviation of current estimation from previous estimations to enforce continuity in time domain, including determining errors between speed estimation and at least one of a speed prediction, a standard score, or a Mahalanobis distance to determine an estimation correctness and to modify the adaptive weight w if incorrect estimation is determined.


 
2. The method of claim 1, wherein modifying the adaptive weight w includes modifying the adaptive weight w in real time.
 
3. The method of claim 1, wherein modifying the adaptive weight includes a comparing a Mahalanobis distance D to a threshold T, and setting a value for the adaptive weight w such that,




 
4. A speed estimation system for a rotational system, comprising:

a vibrational sensor configured to output vibrational signals;

a speed estimation module operatively connected to the vibrational sensor to receive the vibrational signals and output estimated speed data based on the vibrational signals; and

an adaptively weighted filter module configured to receive the estimated speed data and configured to output filtered estimated speed data,

wherein the adaptively weighted filter module includes a Kalman filter having an adaptive weight, the Kalman filter including a Kalman gain Kk, wherein

wherein Pk- is an a priori prediction error covariance, R is a covariance matrix of measurement noises, and w is the adaptive weight;

wherein filtering the estimated speed data through the Kalman filter includes using a test statistic to measure the deviation of current estimation from previous estimations to enforce continuity in time domain, including determining errors between speed estimation and at least one of a speed prediction, a standard score, or a Mahalanobis distance to determine an estimation correctness and to modify the adaptive weight w if incorrect estimation is determined.


 
5. The system of claim 4, wherein the system forms part of or is operatively connected to a controller which controls one or more inputs to the rotational system to provide feedback to the controller.
 


Ansprüche

1. Verfahren zum Schätzen einer Drehgeschwindigkeit eines Systems, umfassend:

Empfangen von Vibrationsdaten von einem Sensor;

Schätzen einer Geschwindigkeit aus den Vibrationsdaten, um geschätzte Geschwindigkeitsdaten zu erzeugen; und

Filtern der geschätzten Geschwindigkeitsdaten über einen adaptiv gewichteten Filter, um inkorrekte Geschwindingkeitsschätzung zu minimieren;

wobei das Filtern der geschätzten Geschwindigkeitsdaten über den adaptiv gewichteten Filter Filtern der geschätzten Geschwindigkeitsdaten über einen Kalman-Filter beinhaltet, der eine Kalman-Verstärkung Kk beinhaltet, wobei

wobei Pk- eine A-priori-Vorhersagefehlerkovarianz ist, R eine Kovarianzmatrix von Messrauschen ist und w die adaptive Gewichtung ist;

wobei das Filtern der geschätzten Geschwindigkeitsdaten über den Kalman-Filter Verwenden einer Teststatistik beinhaltet, um die Abweichung einer Ist-Schätzung von vorherigen Schätzungen zu messen, um Kontinuität in der Zeitdomäne durchzusetzen, darunter Bestimmen von Fehlern zwischen Geschwindigkeitsschätzung und zumindest einem von einer Geschwindigkeitsvorhersage, einem Standardwert oder einer Mahalanobis-Distanz, um eine Korrektheit der Schätzung zu bestimmen und die adaptive Gewichtung w zu modifizieren, falls eine inkorrekte Schätzung bestimmt wird.


 
2. Verfahren nach Anspruch 1, wobei das Modifizieren der adaptiven Gewichtung w Modifizieren der adaptiven Gewichtung w in Echtzeit beinhaltet.
 
3. Verfahren nach Anspruch 1, wobei das Modifizieren der adaptiven Gewichtung ein Vergleichen einer Mahalanobis-Distanz D mit einem Schwellenwert T und Einstellen eines Wertes für die adaptive Gewichtung w beinhaltet, sodass




 
4. Geschwindigkeitsschätzungssystem für ein Drehsystem, umfassend:

einen Vibrationssensor, der dazu konfiguriert ist, Vibrationssignale auszugeben;

ein Geschwindigkeitsschätzungsmodul, das mit dem Vibrationssensor wirkgekoppelt ist, um die Vibrationssignale zu empfangen und geschätzte Geschwindigkeitsdaten auf Grundlage der Vibrationssignale auszugeben; und

ein adaptiv gewichtetes Filtermodul, das dazu konfiguriert ist, die geschätzten Geschwindigkeitsdaten zu empfangen, und dazu konfiguriert ist, gefilterte geschätzte Geschwindigkeitsdaten auszugeben,

wobei das adaptiv gewichtete Filtermodul einen Kalman-Filter mit einer adaptiven Gewichtung beinhaltet, wobei der Kalman-Filter eine Kalman-Verstärkung Kk beinhaltet, wobei

wobei Pk- eine A-priori-Vorhersagefehlerkovarianz ist, R eine Kovarianzmatrix von Messrauschen ist und w die adaptive Gewichtung ist;

wobei das Filtern der geschätzten Geschwindigkeitsdaten über den Kalman-Filter Verwenden einer Teststatistik beinhaltet, um die Abweichung einer Ist-Schätzung von vorherigen Schätzungen zu messen, um Kontinuität in der Zeitdomäne durchzusetzen, darunter Bestimmen von Fehlern zwischen Geschwindigkeitsschätzung und zumindest einem von einer Geschwindigkeitsvorhersage, einem Standardwert oder einer Mahalanobis-Distanz, um eine Korrektheit der Schätzung zu bestimmen und die adaptive Gewichtung w zu modifizieren, falls eine inkorrekte Schätzung bestimmt wird.


 
5. System nach Anspruch 4, wobei das System einen Teil einer Steuerung, die einen oder mehrere Eingänge für das Drehsystem steuert, bildet oder damit wirkverbunden ist, um Feedback für die Steuerung bereitzustellen.
 


Revendications

1. Procédé d'estimation de vitesse de rotation d'un système, comprenant :

la réception de données de vibration d'un capteur ;

l'estimation d'une vitesse à partir des données de vibration pour créer des données de vitesse estimées ; et

le filtrage des données de vitesse estimées à travers un filtre pondéré de manière adaptative afin de minimiser une estimation de vitesse incorrecte ;

dans lequel le filtrage des données de vitesse estimées à travers le filtre pondéré de manière adaptative comporte le filtrage des données de vitesse estimées à travers un filtre

de Kalman qui comporte un gain de Kalman Kk, dans lequel
dans lequel Pk- est une covariance d'erreur de prédiction a priori, R est une matrice de covariance de bruits de mesure, et w est la pondération adaptative ;

dans lequel le filtrage des données de vitesse estimées à travers le filtre de Kalman comporte l'utilisation d'une statistique de test pour mesurer l'écart de l'estimation actuelle par rapport à des estimations précédentes afin d'assurer la continuité dans le domaine temporel, y compris la détermination d'erreurs entre l'estimation de vitesse et au moins l'un parmi une prédiction de vitesse, un score standard, ou une distance de Mahalanobis pour déterminer l'exactitude d'une estimation et modifier la pondération adaptative w si une estimation incorrecte est déterminée.


 
2. Procédé selon la revendication 1, dans lequel la modification de la pondération adaptative w comporte la modification de la pondération adaptative w en temps réel.
 
3. Procédé selon la revendication 1, dans lequel la modification de la pondération adaptative comporte la comparaison d'une distance de Mahalanobis D à un seuil T, et la définition d'une valeur pour la pondération adaptative w telle que




 
4. Système d'estimation de vitesse pour un système de rotation, comprenant :

un capteur de vibration configuré pour émettre des signaux de vibration ;

un module d'estimation de vitesse relié de manière opérationnelle au capteur de vibration pour recevoir les signaux de vibration et produire des données de vitesse estimées sur la base des signaux de vibration ; et

un module de filtre pondéré de manière adaptative configuré pour recevoir les données de vitesse estimées et configuré pour produire des données de vitesse estimées filtrées,

dans lequel le module de filtre pondéré de manière adaptative comprend un filtre de Kalman ayant une pondération adaptative, le filtre de Kalman comportant un gain de Kalman Kk, dans lequel

dans lequel Pk- est une covariance d'erreur de prédiction a priori, R est une matrice de covariance de bruits de mesure, et w est la pondération adaptative ;

dans lequel le filtrage des données de vitesse estimées à travers le filtre de Kalman comporte l'utilisation d'une statistique de test pour mesurer l'écart de l'estimation actuelle par rapport à des estimations précédentes afin d'assurer la continuité dans le domaine temporel, y compris la détermination d'erreurs entre l'estimation de vitesse et au moins l'un parmi une prédiction de vitesse, un score standard, ou une distance de Mahalanobis pour déterminer l'exactitude d'une estimation et modifier la pondération adaptative w si une estimation incorrecte est déterminée.


 
5. Système selon la revendication 4, dans lequel le système fait partie de ou est relié de manière opérationnelle à un dispositif de commande qui commande une ou plusieurs entrées du système de rotation afin de fournir une rétroaction au dispositif de commande.
 




Drawing











Cited references

REFERENCES CITED IN THE DESCRIPTION



This list of references cited by the applicant is for the reader's convenience only. It does not form part of the European patent document. Even though great care has been taken in compiling the references, errors or omissions cannot be excluded and the EPO disclaims all liability in this regard.

Patent documents cited in the description




Non-patent literature cited in the description