An engine fitted to a single-engine aircraft is generally more constrained on safety than in the context of a twin-engine application. In fact, for a twin-engine application, an engine shutdown is an event with less criticality than for a single-engine application where the event is classified as very dangerous since a loss of single engine thrust results in the loss of the propulsion of the engine. 'aircraft.

Failure management for such an application therefore needs to be adapted in order to avoid unacceptable thrust losses as much as possible (called LOTO: Lost Of Thrust Control). Acquisitions that can generate a LOTO are generally physically redundant (duplex acquisitions) in order to prevent a simple electrical failure leading to LOTO (CS-E 50 certification requirement for example).

The mechanical or electrical or electronic system or any combination involved in duplex acquisition is therefore redundant: it comprises at least two channels A and B performing the same acquisition function.

Physical redundancy makes it possible to detect simple failures and to adapt to them. Indeed, a difference test D between channels A and B (for example the calculation of the difference between the values ​​measured by each channel A, B, such as an intensity or a voltage, or their corresponding digital signals) is

generally performed on acquisitions in the control program to detect these simple failures.

Figure 1 illustrates this method. It will be described later.

This difference test D calculates the difference D between channels A, B and compares it with a predefined detection threshold Th_D:

If the difference D between channels A and B is less than the detection threshold Th_D, it is considered that there is no fault (either the two channels A, B are indeed healthy, or one channel is slightly faulty but the failure is small enough to have no impact on the engine),

If the difference D between the channels A and B is greater than the detection threshold Th_D, it is considered that there is a failure on one of the two channels A, B since this difference D is too large to reflect a real physical behavior of the motor. It is then said that a failure is detected.

Once detected, it is then necessary to locate the fault, that is to say to determine the faulty channel among the two tested channels.

Generally, the localization is carried out simultaneously with the detection by making a three-way vote with an arbitration model: using a model of the measured physical quantity (the model providing modeled values ​​of the physical quantity which have been developed from measurements of other physical quantities in the engine environment), channel A, B closest to this model, i.e. channel A, B whose acquired value is closest to the modeled value of the physical quantity is considered valid.

In other words, when a failure is detected and channel A (respectively B) is close to the model, then channel B (respectively A) is considered to be faulty and channel A (respectively B) is selected. For example, this selection means that it will be channel A which will be used for all the other actions concerning the aircraft. Channel B is functionally ignored.

The problem with this strategy is that the accuracy of the model is generally much lower than the accuracy of the sensor which can quickly lead to poor locations.

The example below will illustrate this situation.

In practice, the value selected for the various processing operations subsequent to acquisition (use of the measurements acquired in other calculations, etc.) is generally the arithmetic mean between channels A, B in nominal operation (ie when the two channels A and B are healthy), which then makes it possible to halve the impact of a failure of one of the two channels on the selected value (in FIG. 1, the dotted line up to t = t1). However, in the event of poor localization (that is to say that if, in the event of a failure, the faulty channel is selected), the error is taken in full, which is the most unfavorable case.

Indeed, if we consider that Voie_saine = Real_value and

Default_path = Real_Value + Failure, it can be seen that the selection of the defective path is more unfavorable than the absence of accommodation.

If we choose the healthy path, then we have

Selected_value = Healthy_Channel = Actual_Value

If we choose the mean between channels, then we have

Selected_value = 1/2 (Healthy_Channel + Faulty_Channel)

= 1/2 (Actual value + Actual value + Failure)

= Actual_value + 1/2 Failure

If we choose the faulty channel, then we have

Selected_value = Faulty_channel = Actual_value + Fault

Thus, in the event of a bad selection, the failure is suffered in its entirety, against only half of the failure in the case of taking the average into account.

In the example of FIG. 1, we consider a sensor with a drifting channel B. It is assumed that a feared event (LOTO for example) occurs if the 8K is wrong on the measurement (failure threshold Th Err).

We define (design parameter) a detection threshold of the difference between channels (“cross-check” threshold) Th_D of 1 K and we impose that the localization be done at the same time as the detection. In addition, in nominal operation, the average between channels A, B is selected.

At time t0, one of the two lanes drifts (suppose it is lane B). The selected curve, which corresponds to the average at nominal speed, drifts twice as slowly (in figure 1, the dotted line up to t = t1). At t1, channel B reaches the detection threshold. As the sensors of channels A and B are precise, this detection is reliable (ie there is really a problem).

The localization must also be done at t1, it is necessary to compare the channels A and B with the model. On the other hand, as the detection took place very early (with a low D deviation), the “bad” precision of the model does not always make it possible to distinguish the faulty channel from the healthy channel. Thus, if the selection is made on channel B, the selected curve B will reach the threshold of the event Th Err more quickly than if the selected curve had remained that of the average of channels A, B. We therefore rejected the correct one. lane A and the situation immediately deteriorates.

It is then observed that in certain cases, a too hasty localization leads to amplify a phenomenon which would have been weaker without action.

It would therefore be better to set the right time to locate in order to minimize the likelihood of poor localization.

Finally, another approach sometimes used is not to detect a fault until one is able to locate it. The disadvantage in this case is to delay the detection of the failure.

The invention proposed below defines a strategy which makes it possible to greatly reduce the number of false locations and therefore leads to more reliable regulation systems.

PRESENTATION OF THE INVENTION

The invention starts from the observation that the bad localization rate is relatively high with the current failure management strategy. Indeed, fault detection and localization are carried out simultaneously although these two phases do not meet the same need:

detection is an operation essentially based on the precision of the sensors and the detection threshold is generally determined by this precision, the localization is an operation based on the precision of the arbitration model, during which one seeks to evaluate the most likely path true based on the model output.

The detection and localization phases therefore have every interest in being distinguished because they are not directed by the same sizing parameters. When a failure is of relatively low amplitude, it is safer not to accommodate it (conservation of the average between channels) than to seek to locate it while taking a significant risk of making a mistake.

In this regard, according to a first aspect, the invention proposes a method for detecting and locating a fault in an acquisition system comprising two redundant acquisition channels for the measurement of a physical quantity in an environment, the method using a calculation unit comprising a memory storing a model of the measured physical quantity, said model providing modeled values ​​of the physical quantity as a function of measurements of other physical quantities in said environment, the method implementing the following steps:

- (E1) detection of a symptomatic error of a defective acquisition channel when a difference between the measured values ​​of the two channels (A, B) reaches a detection threshold,

- (E2) waiting to let the acquisition system evolve for a certain period of time to be kept under control of the error,

- (E3) location of the defective channel among the two channels, when the difference in the values ​​measured between the channels reaches a location threshold, said location being carried out from the comparison of the measured value of each of the channels with a modeled value of physical magnitude,

the localization threshold being different from the detection threshold.

The duration T is preferably non-zero. It corresponds to a duration of inoffensiveness of the error.

In one embodiment, the calculation unit generates work data which takes into account the values ​​of the two acquisition channels as long as the location step (E3) is not triggered, said method comprising a step of :

- (E4) selection of the non-defective channel as the working quantity once the step of locating (E3) the defective channel has been carried out.

In one embodiment, a failure threshold is predefined and the calculation unit generates work data which takes into account the values ​​of the two acquisition channels as long as the location step (E3) is not triggered. , and in which the localization threshold is chosen so that the difference between the values ​​of the two channels can exceed the failure threshold but that the difference between the work data and the value of one or the other of the two-way cannot exceed the failure threshold.

In one embodiment, the location threshold is greater than a failure threshold.

In one embodiment, the working quantity corresponds to an output quantity of the unit for subsequent processing by other computers.

In one embodiment, the working quantity corresponds to an arithmetic mean of the quantities of the channels.

In one embodiment, the location threshold is included (preferably strictly) between the failure threshold and twice the failure threshold.

In one embodiment, a detection notification step is sent by the computing unit before the implementation of the location step (E4), preferably at the time of the detection step (E3), for provide information on a state of failure detected but not located.

In one embodiment, the detection threshold and / or the localization threshold and / or the failure threshold and / or the model of the physical quantity is / are a function of the operating ranges of the acquisition system and / or of the precision of the model.

In one embodiment, the detection threshold is set based on the sensor specifications of the two channels.

In one embodiment, the thresholds and the deviations are expressed in absolute value (and are therefore positive).

According to a second aspect, the invention proposes a calculation unit for the detection and location of failure of a measurement acquisition channel in an acquisition system comprising two redundant acquisition channels for the measurement of a quantity physical quantity in an environment, the computing unit comprising a memory (24) storing a model (Mod) of the measured physical quantity, said model providing modeled values ​​of the physical quantity as a function of measurements of other physical quantities in said environment , the computing unit being configured to receive measurement data from the acquisition channels of the acquisition system,

the computing unit being configured to implement the steps of the method described above, that is to say in particular the following steps:

- (E1) detection of an error when the difference between the measured values ​​of the two channels exceeds a detection threshold,

- (E2) waiting to let the acquisition system evolve for a certain period of time,

- (E3) location of the defective channel among the two channels, when the difference in the values ​​measured between the channels crosses a location threshold, said location being carried out from the comparison of the measured value of each of the channels with a value modeled of the physical quantity,

the localization threshold being different from the detection threshold.

According to a third aspect, the invention proposes an assembly comprising a calculation unit as described above and a duplex acquisition system.

Selon un quatrième aspect, l’invention propose un procédé de localisation de panne dans un système d’acquisition comprenant deux voies d’acquisition redondantes de la mesure d’une grandeur physique dans un environnement, le procédé utilisant une unité de calcul comprenant une mémoire stockant un modèle de la grandeur physique mesurée, ledit modèle fournissant des valeurs modélisées de la grandeur physique en fonction de mesures d’autres grandeurs physiques dans ledit environnement, le procédé mettant en oeuvre les étapes suivantes :

- (E3) localisation de la voie défectueuse parmi les deux voies, lorsqu’un écart des valeurs mesurées entre les voies atteint un seuil de localisation, ladite localisation étant effectuée à partir de la comparaison de la valeur mesurée de chacune des voies avec le modèle de la voie stocké dans la mémoire,

in which a failure threshold is defined and the calculation unit generates work data which takes into account the values ​​of the two acquisition channels as long as the location step (E3) is not triggered, and in which the location threshold is chosen so that the difference D between the values ​​of the two channels can exceed the failure threshold but that the difference between the work data and one or the other of the two channels cannot exceed the failure threshold (preferably the larger deviation of the two deviations).

The localization step here involves the detection step.

According to a fifth aspect, the invention proposes a calculation unit for the detection and location of faults in an acquisition system comprising two redundant acquisition channels for the measurement of a physical quantity in an environment, the unit of calculation comprising a memory storing a model of the measured physical quantity, said model providing modeled values ​​of the physical quantity as a function of measurements of other physical quantities in said environment, the calculation unit being configured to receive measurement data from acquisition channels of the acquisition system, the calculation unit being configured to implement the steps of the method described above, that is to say in particular the following steps:

- (E3) location of the defective channel among the two channels, when a difference in the values ​​measured between the channels reaches a location threshold, said location being carried out from the comparison of the measured value of each of the channels with the model of the channel stored in the memory,

in which a failure threshold is defined and the calculation unit generates work data which takes into account the two acquisition channel values ​​as long as the location step (E3) is not triggered, and in which the localization threshold is chosen so that the difference D between the values ​​of the two channels can exceed the failure threshold but the difference between the work data and one or the other of the two channels cannot exceed the failure threshold (preferably the greater of the two deviations).

PRESENTATION OF FIGURES

Other characteristics, aims and advantages of the invention will emerge from the following description, which is purely illustrative and non-limiting, and which should be read in conjunction with the accompanying drawings, in which:

FIG. 1 illustrates a known detection and localization method.

FIG. 2 illustrates an assembly making it possible to implement the invention.

FIG. 3 illustrates a method according to an embodiment of the invention.

DETAILED DESCRIPTION

With reference to FIG. 2, an assembly comprising a duplex or redundant acquisition system 10 and a calculation unit 20 capable of receiving data originating from the acquisition system 10 has been illustrated.

The acquisition system 10 comprises a first channel A and a second channel B making it possible to measure data relating to a mechanical, electronic, electrical and / or hydraulic device. The channels A, B are intended to measure the same physical quantity of the same device 30. The channels A and B each comprise for this each at least respectively one sensor 12, 14 (voltage, magnetic field, temperature, torque sensor, etc. by force, etc.).

The sensors 12, 14 of the two channels A, B then send their data to the computing unit 20. The data is generally in the form of an electrical voltage value indicative of the parameter that the sensors 12, 14 measure. Data can be transmitted digitally or analog. In particular, the data can be preprocessed (filtering, smoothing, etc., either at the level of the sensor or at the level of the calculation unit 20) in order to be able to be used in the case of a method conforming to the various embodiments. of the invention.

The acquisition system 10 can thus include a processing module 16 for carrying out the pre-processing of the data coming from the channels A, B so that the calculation unit 20 receives data that can be used directly.

The calculation unit 20 comprises a processor 22 able to process data (for example calculate the differences between the values ​​of the acquisition channels A, B, calculate a new quantity from these acquisition channels A, B, calculate an average, etc.) and perform calculations, and further comprises a memory 24. The processor 22 can communicate with the memory 24.

The calculation unit 20 also has the function of generating work data S, which is then used by the other computers of the aircraft. The work data S corresponds to a function of the values ​​of the acquisition channels A, B, which can be written in the form S = f (A, B). The principle of redundancy calls for a principle of symmetry in the function f (because the two paths A, B are generally preferably undifferentiated and neither is privileged over the other) so we preferably choose a function f such that f (A, B) = f (B, A). This may not be the case: sometimes, for certain temperatures, it is preferable to favor the maximum between channels to protect against over-temperatures.

Dans un mode de réalisation préférentielle, la fonction f est une moyenne et on a S=moy_AB=(A+B)/2. Cela signifie que la donnée de travail S correspond à la moyenne des valeurs des voies d’acquisition A et B.

La mémoire 24 stocke un modèle informatique Mod de la grandeur mesurée par les voies A, B. Ce modèle a été élaboré préalablement à partir d’autres mesures.

Comme indiqué précédemment, la précision des capteurs 12, 14 est meilleure que celle du modèle Mod.

L’unité de calcul 20 peut regrouper plusieurs sous-unités de calcul, notamment dans le cas où les tâches sont partagées entre différents calculateurs ou ordinateurs.

Les voies d’acquisition A, B acquièrent en continu (ou à intervalle régulier) des données relatives au dispositif 30. Ces données sont envoyées à l’unité de calcul 20 qui calcule notamment un écart D entre les valeurs mesurées des voies A, B. Cet écart D peut être en valeur absolue ou en relatif (positive ou négative). Pour illustrer certaines situations dans la présente description, on considérera que les valeurs acquises vérifient la définition de D suivante : B=A+D.

On définit un seuil de défaillance Th_Err à partir duquel on considère que l’écart D sur la mesure, consécutif à une panne, risque de provoquer un évènement redouté ou indésiré (LOTC par exemple). Cela signifie que si la donnée de travail S correspond à la voie d’acquisition défectueuse (ce que l’on souhaite éviter donc), il y a un risque de provoquer l’évènement indésiré.

With reference to FIG. 3, which graphically illustrates the different steps of the detection and location method, this method will be described. We assume here a failure threshold of 8K.

In a first step E1, an error, consecutive to a failure, is detected when the deviation D reaches a detection threshold Th_D (1 K in FIG. 3). The error is symptomatic of a faulty acquisition channel. The data relating to the detection threshold Th_D is typically stored in the memory 24. The detection threshold Th_D depends on the acquisition system 10 including in particular the specifications of the sensors 12, 14 (therefore on their precision).

Step E1 is implemented by the calculation unit 20. In FIG. 3, step E1 occurs at t1.

Optionally, a notification step E1 ′ is triggered by the calculation unit 20 to provide information according to which an error has been detected. The error detection information can be used by a computer, another module of the computing unit 20, or even by a human operator, etc.). Consequently, there is error detection information relating to the acquisition system indicating that one of the two channels A, B is defective, but without being able at this stage to identify which one.

Then, a waiting step E2 is performed, which simply corresponds to a wait to let the system evolve for a period T. During this waiting step E2, the failure is known but under control, that is to say. say that we know that the generated error will not lead to the feared event: it is therefore a time of keeping the error under control or a duration of inoffensiveness of the error . The duration T lasts until a step E3 for locating the defective channel is triggered (see FIG. 3). The duration T is typically non-zero (except in the case of clear failure, where a channel moves instantaneously).

The location step E3 is performed when the difference D of the values ​​measured between the channels A, B reaches a location threshold Th_L (FIG. 3). The localization is done by comparing the measured values ​​for channels A, B with the Mod model of channels A, B stored in memory 24. Typically, channel A, B considered as non-defective is the one which is the closest to the model. Mod. Closest means that the difference between the values ​​is the smallest.

Step E3 is implemented by the calculation unit 20. In FIG. 3, step E3 occurs at t2.

Thanks to the wait E2, the difference D has been able to grow since the detection step E1 (for example in the case of a sensor gain drift), which means that the identification of the channel is done with less risk of error despite the imprecision of the model Mod.

Naturally, the localization threshold Th_L has a value different from that of the detection threshold Th_D, otherwise the same risks of error linked to an early localization would be present.

In addition, as the detection E1 has been carried out upstream, a new state of failure detected but failure not located is created. There is thus a preventive detection which can cause the implementation of certain preventive measures: not taking into account the data generated by the acquisition system 10 or maintenance operation. Indeed, even if the risk of using the wrong channels A, B is high at this stage, the criticality in maintenance is much less than during operation. It is therefore reasonable in this context to try to locate the error earlier.

Once the location E3 has been performed and the defective channel has been identified (or concomitantly), a selection step (also called accommodation) E4 of the non-defective channel is implemented. This means that the work data S of the calculation unit 20 becomes the non-defective acquisition channel - the other channel being ignored.

The selection step E4 is implemented by the calculation unit 20. In FIG. 3, the step E4 also occurs at t1.

There are therefore two thresholds to be adjusted: the detection threshold Th_D and the localization threshold Th_L. The Th_D detection threshold depends only on the sensor specifications.

On the other hand, it remains to determine the value of the localization threshold Th_L, which imposes to know the moment from which one would end up with the feared event without accommodation.

In practice, as was indicated earlier, the work data S corresponds to the average Moy_AB of the acquisition channels A, B. By definition, if we are wrong about the measurement of a deviation D equal to Th_Err and the work data S corresponds to the defective channel, there is a risk of ending up with the feared event.

However, as long as the defective acquisition channel has not been located, the failure has an impact divided by two (because the work data S is the average Moy_AB between the two channels A, B).

Therefore, the risk only occurs when SA = Moy_AB-A = Th_Err-A. We therefore have Th_Err-A = (A + B) / 2-A = (A + A + D) / 2-A = D / 2, that is to say A = 2Th_Err.

It is therefore possible to implement the location step E3 only when the difference D between the channels reaches 2Th_Err, that is to say as late as possible. This value being extreme, the application of a safety margin means that one seeks to locate when the difference D between channels reaches Th_L such that Th_Err Th_Err (Th_L is strictly greater than Th_Err), the difference D between channels A, B will be caused to exceed the failure threshold Th_Err during the waiting step E2. On the other hand, as Th_L £ 2Th_Err, the working data S = Moy_AB does keep a gap with channel A (or channel B for that matter, except for the sign, because the function is the mean function) less than Th Err during any the waiting phase E2, which means that the risk that the feared event

occurs is avoided. If Th_L = 2Th_Err, the localization step is triggered at the moment when the work data S = Moy_AB reaches a difference with the channel A which is worth Th Err. If Th_L <2Th_Err, then the work data S = Moy_AB does keep a gap with channel A which will always be less than Th Err.

As other output data S are possible (use of functions f other than the mean), the principle can be immediately generalized: a localization threshold Th_L is chosen so that the difference between the values ​​of the two channels A, B can exceed the failure threshold Th Err but that the difference between the work data S and one or the other of the two channels A, B (we can calculate the two differences and take the maximum of the differences) cannot exceed the failure threshold Th Err.

In this way, we locate “at the latest” when the error on the faulty acquisition channel is maximum, which greatly reduces the probability of bad localization.

The values ​​of the thresholds Th Err, Th_L_, Th_D are stored in the memory 24 of the calculation unit 20. The thresholds are here expressed in absolute value, that is to say they are positive.

As specified above, the thresholds and the differences are advantageously expressed in absolute value, so as not to depend on the signs of the data acquired. However, it is possible to work outside the absolute value: it is then necessary to take into account the signs and the evolutions (increasing or decreasing) of the data.

By virtue of the method described, the maximum tolerances are used as much as possible on the acquisition before the location step E3 to accommodate as needed. Consequently, the more this criticality is controlled, the more the performance of failure management can be improved.

Finally, the operation of the acquisition system 10 and / or the precision of the model Mod may vary depending on the operating regime of the device 30. In this case, different values ​​of each threshold Th Err, Th_L and / or Th_D , and / or the Mod model can be stored in a table in memory

24, depending on the operating ranges of the acquisition system 10 and / or the accuracy of the model.

CLAIMS

1. Method of detecting and locating a failure of a measurement acquisition channel in an acquisition system (10) comprising two redundant acquisition channels (A, B) of the measurement of a physical quantity in an environment, the method using a computing unit (20) comprising a memory (24) storing a model (Mod) of the measured physical quantity, said model providing modeled values ​​of the physical quantity as a function of measurements of other physical quantities in said environment , the method implementing the following steps:

- (E1) detection of a symptomatic error of a defective acquisition channel when a difference (D) between the measured values ​​of the two channels (A, B) reaches a detection threshold (Th_D),

- (E2) waiting to let the acquisition system evolve for a certain period (T) of keeping the error under control,

- (E3) location of the defective channel among the two channels (A, B), when the difference (D) of the values ​​measured between the channels (A, B) reaches a location threshold (Th_L), said location being performed from the comparison of the measured value of each of the channels (A, B) with a modeled value of the physical quantity,

the localization threshold (Th_L) being different from the detection threshold (Th_D).

2. Method according to claim 1, wherein the calculation unit (12) generates a working data item (S, moy_AB) which takes into account the values ​​of the two acquisition channels (A, B) as long as the step location (E3) is not triggered, said method comprising a step of:

- (E4) selection of the non-defective channel (B, A) as the working quantity (S) once the step of locating (E3) of the defective channel (A, B) has been carried out.

3. Method according to any one of claims 1 to 2, in which a failure threshold (Th Err) is predefined and the calculation unit (12) generates a work datum (S, moy_AB) which takes into account the values ​​of the two acquisition channels (A, B) as long as the localization step (E3) is not triggered, and in which the localization threshold (Th_L) is chosen so that the difference (D) between the values ​​of the two channels (A, B) may exceed the failure threshold (Th Err) but the difference between the work data (Moy_AB) and the value of one or the other of the two channels (A, B) cannot exceed the failure threshold (Th Err).

4. The method of claim 3, wherein the location threshold (Th_L) is greater than a failure threshold (Thr Err).

5. Method according to any one of claims 1 to 4, wherein the work quantity (moy_AB) corresponds to an arithmetic mean of the quantities of the channels (A, B).

6. The method of claim 5, wherein the location threshold (Th_L) is between the failure threshold (Th Err) and twice the failure threshold (Th_Err).

7. Method according to any one of claims 1 to 6, wherein a notification step (AND) of the detection is sent by the calculation unit (10) before the implementation of the location step (E4 ), preferably at the time of the detection step (E3), to provide information on a fault state detected but not located.

8. Method according to any one of claims 1 to 7, wherein the detection threshold (Th_D) and / or the localization threshold (Th_L) and / or the failure threshold (Th Err) and / or the model of the physical quantity (Mod) is / are a function of the operating ranges of the acquisition system (10) and or of the precision of the model.

9. Method according to any one of claims 1 to 8, wherein the detection threshold (Th_D) is set according to the sensor specifications of the two channels (A, B).

10. Calculation unit (12) for the detection and location of failure of a measurement acquisition channel in an acquisition system (10) comprising two redundant acquisition channels (A, B) of the measurement d 'a physical quantity in an environment, the calculation unit comprising a memory (24) storing a model (Mod) of the measured physical quantity, said model providing values

modeled of the physical quantity as a function of measurements of other physical quantities in said environment, the computing unit being configured to receive measurement data from the acquisition channels (A, B) of the acquisition system (10) ,

the calculation unit (12) being configured to implement the following steps:

- (E1) detection of an error of a symptomatic error of a defective acquisition channel when the difference between the measured values ​​of the two channels (A, B) crosses a detection threshold (Th_D),

- (E2) waiting to let the acquisition system evolve for a certain period (T) of keeping the error under control,

- (E3) location of the defective channel among the two channels (A, B), when the difference in the values ​​measured between the channels (A, B) crosses a location threshold (Th_L), said location being performed from the comparison of the measured value of each of the channels (A, B) with a modeled value of the physical quantity,

the localization threshold (Th_L) being different from the detection threshold (Th_D).