DETAILED ACTION
This Office Action is taken in response to Applicant’s Amendment and Remarks filed on 6/30/2025
regarding Application No. 17/941,597 originally filed on 09/09/2022. Claims 1-15 as filed are
currently pending and have been considered as follows:
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Arguments
The applicant argues “it is respectfully submitted that Shattil does not particularly disclose ways in which qualities representative of forces exerted on control surfaces can be used to calculate an estimated speed of the aircraft. In this regard, it is respectfully submitted that the amendments to claim 1 discussed above distinguishes the general operation of a neural network taught by Shattil by requiring that the calculation of the at least one speed of the aircraft” [Remarks, p. 9]. The examiner respectfully disagrees.
Goupil expressly discloses using quantities representative of forces on aircraft control surfaces (aerodynamic hinge moment) together with other flight data to compute speed on-board, (as per “determining at least one aerodynamic hinge moment of at least one control surface… determining… a static pressure… and a model of hinge moment coefficient; [and] computing a Mach number… representing a first speed of the aircraft” in ¶6-¶18). Shattil teaches a feed-forward artificial neural network that maps a set of input data (received from sensors) to a set of output data, with training/updates to minimize error (¶7-¶9, ¶22, ¶41-¶47). Shattil further discloses that inputs “can comprise measurements of physical signals… received from at least one sensor” (¶22) and that the ANN “maps a set of input data onto a set of output data” (¶7).
Accordingly, it would have been obvious to one of ordinary skill before the effective filing date of the invention to use Shattil’s ANN as an alternative computational scheme within Goupil’s framework to perform the mapping that Goupil requires. Goupil further disclsoes that the hinge-moment/speed relationship may be complex and “non-invertible,” and expressly teaches alternative estimation schemes (e.g., Least-Squares Sliding-Mode, Luenberger, high-order sliding mode) (¶128-¶129, ¶169-¶177, ¶180-¶189). Substituting the ANN as taught by Shattil for Goupil’s listed observers is a predictable choice to determine the same nonlinear mapping using available sensor inputs to produce the same output (speed). Applicant’s traversal is therefore unpersuasive.
The applicant argues “it is respectfully submitted that the general disclosure by Shattil of a neural network that includes an input layer and an output layer is not equivalent to the claimed neural network in which the at least one speed to be calculated corresponds to an output layer of the neural network and in which the pressure difference and the plurality of flight parameters correspond to an input layer of the neural network.” [Remarks, p. 9]. The examiner respectfully disagrees.
Goupil discloses of the specific inputs and output: inputs include (“measuring at least one pressure difference between two hydraulic chambers… determining… static pressure,” and additional flight parameters such as “aerodynamic angles… angle of deflection… load factors… aerodynamic configuration… angular velocities” (¶20, ¶29, ¶129, ¶135). The output is “a Mach number… representing a first speed of the aircraft,” (¶6, ¶36, ¶81-¶87). Shattil discloses of the ANN topology and mapping: inputs provided to an input layer; outputs produced at an output layer; and the ANN “maps a set of input data onto a set of output data” ¶7, ¶41-¶47. Shattil also teaches using sensor-measured signals as inputs (¶22). Goupil supplies the specific input signals and the specific output; Shattil supplies the ANN implementing an input layer receiving measured signals and an output layer producing the estimated speed. Applicant’s traversal is therefore unpersuasive.
The applicant argues “it is respectfully submitted that a person having ordinary skill in the art would not understand this particular configuration as being obvious in view of the general teaching of a feed-forward neural network provided by Shattil.” [Remarks, p. 10]. The examiner respectfully disagrees.
It would have been obvious before the effective filing date to combine Goupil with Shattil to enable another standard means of performing the nonlinear estimation that Goupil already discusses. Goupil explicitly acknowledges the non-invertibility of the hinge-moment model and proposes multiple observer-based algorithms (e.g., Least-Squares Sliding-Mode, Luenberger, high-order sliding mode) (¶128-¶129, ¶169-¶177, ¶180-¶189). Shattil discloses well-known ANN techniques for mapping sensor inputs to target outputs (¶6-¶11). Implementing Goupil’s estimation with Shattil’s ANN would have predictably yielded the same type of result (on-board real-time speed estimate without Pitot), with potential benefits expressly taught by Shattil such as computational efficiency via structured updates, adaptability, and robustness (¶6, ¶9, ¶11, ¶21). Accordingly, the proposed combination would have been obvious to a POSITA, and applicant’s argument is unpersuasive.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim(s) 1-2, 5-6, 8-9, 11-14 are rejected under 35 U.S.C. 103 as being unpatentable over Goupil (US Pub. No. 20160107762) in view of Shattil (US Pub. No. 20200364545).
As per Claim 1, Goupil discloses a system for automatically estimating a speed of an aircraft during a flight of the aircraft (as per Abstract), comprising:
a plurality of control surfaces of the aircraft, each of the plurality of control surfaces comprising an actuator configured to deflect a corresponding one of the plurality of control surfaces of the aircraft; (as per “the device 1 comprises a set 11 of units 2 for determining aerodynamic hinge moments of a plurality of different control surfaces 3 of the aircraft, and the central unit 5 computes the Mach number” in ¶87, as per “the deflection of this control surface 3 is generated by one or more actuators 12 which can be of any type, in particular of hydraulic type, of electro-hydrostatic type or of electromechanical type” in ¶79)
a determining module in communication with the plurality of control surfaces of the aircraft, the determining module being configured to determine, for each of the plurality of control surfaces of the aircraft, (as per “a data generating unit 2 configured to determine at least one aerodynamic hinge moment of at least one control surface 3 of the aircraft;” in ¶62-¶64, as per “This aerodynamic hinge moment illustrates the aerodynamic forces acting on the control surface of the aircraft in the course of the flight” in ¶19) at least one quantity which is representative of a force exerted on the corresponding one of the plurality of control surfaces of the aircraft, (as per “measuring at least one pressure difference between two hydraulic chambers of at least one actuator intended to generate the deflection of the control surface; and computing the aerodynamic hinge moment, using at least the pressure difference” in Claim 2),
the at least one quantity which is representative of a force corresponding to a pressure difference measured between two hydraulic chambers of the actuator which can deflect the corresponding one of the plurality of control surfaces of the aircraft; (as per “the device 1 comprises a set 11 of units 2 for determining aerodynamic hinge moments of a plurality of different control surfaces 3 of the aircraft, and the central unit 5 computes the Mach number” in ¶87, as per “the deflection of this control surface 3 is generated by one or more actuators 12 which can be of any type, in particular of hydraulic type, of electro-hydrostatic type or of electromechanical type” in ¶79)
a calculating module configured to calculate at least one speed of the aircraft at least based on the at least one quantity corresponding to each of the plurality of control surfaces determined by the determining module; (as per “computing a Mach number M representing a first speed of the aircraft,” in ¶6-¶18, as per “the speed of the aircraft is estimated automatically, in real time, on the basis of the value of at least one aerodynamic hinge moment of at least one control surface of the aircraft, without using any measurement of the total air pressure, thereby making it possible to dispense with the associated sensors, generally redundant on civil airplanes.” in ¶18)
a transmitting module configured to transmit the at least one speed of the aircraft which is calculated by the calculating module to a user device; (as per “at least one data transmission unit configured to provide the first speed to at least one user system,” in ¶38)
a collecting module configured to collect a plurality of flight parameters from the aircraft; (as per “b) determining a plurality of data and at least the following data: a static pressure external to the aircraft; and a model of hinge moment coefficient;” in ¶6-¶18)
the at least one speed of the aircraft being calculated by the calculating module based on the pressure difference and the plurality of flight parameters using a function in which the at least one speed is a function of the pressure difference (as per “the following expression: Mₐ = ½ [γ · Pₛ · M² · Vδₚ · Cₕ]… Ma is the aerodynamic hinge moment; γ is an adiabatic coefficient of the air; Ps is an external static pressure; Vδ p represents a volume of the control surface; and Ch is the model of aerodynamic hinge moment coefficient of the control surface, dependent on the Mach number M;” in ¶6-¶18)
the function having a following expression:
VCAS = f(θ)
in which VCAS corresponds to the at least one speed to be determined, (as per “the method comprises an additional step, posterior to step c) and consisting in or comprising computing a conventional speed (or corrected speed) CAS, representing a second speed of the aircraft, on the basis of the Mach number computed in step c) and of additional data.” in ¶36)
θ corresponds to a parameter vector comprising the pressure difference determined for each of the plurality of control surfaces of the aircraft by the determining module and the plurality of flight parameters measured by the collecting module; (as per “the following expression: Mₐ = ½ [γ · Pₛ · M² · Vδₚ · Cₕ]… Ma is the aerodynamic hinge moment; γ is an adiabatic coefficient of the air; Ps is an external static pressure; Vδ p represents a volume of the control surface; and Ch is the model of aerodynamic hinge moment coefficient of the control surface, dependent on the Mach number M;” in ¶6-¶18)
Goupil fails to expressly disclose:
the at least one speed being calculated based on a neural network in which the at least one speed to be calculated corresponds to an output layer of the neural network and
in which the pressure difference and the plurality of flight parameters correspond to an input layer of the neural network,
the neural network comprising fixed synaptic weights which are determined off-line through training on data sets which characterize a correspondence between the vector parameter and the at least one speed determined for a plurality of flights of the aircraft.
Shattil discloses of an artificial neural network (as per Abstract), comprising:
the at least one speed being calculated based on a neural network in which the at least one speed to be calculated corresponds to an output layer of the neural network; (as per “layer l might be the input layer, and z0 l can be the ANN input value, x0. In one example, layer l might be the output layer, and a0l can be the ANN output value y0..” in ¶34,)
in which the pressure difference and the plurality of flight parameters correspond to an input layer of the neural network, (as per “The disclosed ANNs can include any type of sensor(s) for providing input data to the ANNs. Sensors can comprise … and/or any type of situational awareness sensors and systems (e.g., accelerometer, inclinometer, inertial sensor, orientation sensor, speedometer, altimeter, primary flight displays (PFDs), flight management systems (FMS), collision avoidance systems (CAS), air traffic control (ATC) systems, navigation systems (e.g., pointing, navigation, and timing receivers), gyroscope, tachometer, oil pressure gauge, fuel gauge, etc.).” in ¶164, as per “data inputs to the ANN can comprise measurements of physical signals, such as Physical-Layer signals used for data communications. The input data may include a signal received in a wireless communication network and/or a signal to be transmitted in the wireless communication network.” in ¶22)
the neural network comprising fixed synaptic weights which are determined off-line through training on data sets which characterize a correspondence between the vector parameter and the at least one speed determined for a plurality of flights of the aircraft. (as per “The weights are then trained (optimized) so that for a given training vector input, the neural network produces an output close to a desired (predetermined) training vector output.” in ¶8, as per “An error for each output neuron, or output node, is then calculated based on the actual neuron output and a target training output for that neuron. Then backpropagation may be performed through the neural network (in a direction from the output layer back to the input layer) to update the weights based on how much effect each weight has on the overall error, so that the output of the neural network moves closer to the desired training output. This cycle can be repeated until the actual output of the neural network is within an acceptable error range of the desired training output.” in ¶8, as per “The error and/or total cost may be used to update parameters in the ANN, such as to reduce the error of the ANN's output, or prediction, such as may be measured from a labeled training data set” in ¶11)
In this way, Shattil operates to update weights and/or biases to update in a learning process (¶6). Like Goupil, Shattil is concerned with aircraft systems. (¶164)
It would have been obvious for one of ordinary skill in the art before the effective filing date to have modified the speed estimation system of Goupil with the neural network system of Shattil to enable another standard means of estimating aircraft speed via a neural network. Such modification also allows the system to compute the gradient of the prediction output, error, or total loss with respect to a change in any of the ANN parameters. (¶152)
As per Claim 2, the combination of Goupil and Shattil teaches or suggests all limitations of Claim 1. Goupil further discloses wherein the determining module comprises at least one collecting submodule configured to collect an individual pressure difference from the actuator or from each actuator, the individual pressure difference being measured by a pressure difference measurement sensor of each actuator. (as per “measuring at least one pressure difference between two hydraulic chambers of at least one actuator intended to generate the deflection of the control surface; and computing the aerodynamic hinge moment, with the aid at least of this pressure difference.” in ¶20-¶22, as per “at least one measurement element configured to measure at least one pressure difference between the hydraulic chambers of at least one actuator intended to generate the deflection of the control surface; and a computation element configured to compute the aerodynamic hinge moment, with the aid at least of this pressure difference.” in ¶50-¶52)
As per Claim 5, the combination of Goupil and Shattil teaches or suggests all limitations of Claim 1. Goupil further discloses of wherein parameter vector θ = [ΔP; PS; α; CONF; p; ẟP], wherein
ΔP corresponds to overall pressure difference or to the individual pressure differences determined by the determining module, (as per “measuring at least one pressure difference between two hydraulic chambers of at least one actuator intended to generate the deflection of the control surface;” in Claim 2)
PS corresponds to static air pressure (as per “P is an external static pressure;” in Claim 1)
α corresponds to an angle of attack of the aircraft (as per “αAC is the angle of attack of the aircraft;” in ¶117)
CONF corresponds to an aerodynamic configuration of slats and flaps of the aircraft (as per “conf represents the aerodynamic configuration of the aircraft, depending in a standard manner on the position of the lift-enhancing devices (leading edge or trailing edge devices, also called slats and flaps);” in ¶119)
p corresponds to an angle of roll of the aircraft. (as per “p, q, and rare the angular velocities of the aircraft, respectively in roll, pitch and yaw.” in ¶120)
ẟP corresponds to an angle of deflection of a control surface or of a position of a shaft of the actuator connected to the control surface of the aircraft. (as per “the angle of deflection ẟP of the control surfaces 3” in ¶131, Fig. 2)
As per Claim 6, the combination of Goupil and Shattil teaches or suggests all limitations of Claim 1. Goupil fails to expressly disclose wherein the neural network has three layers: the input layer, the output layer and a hidden layer between the input layer and the output layer.
See Claim 1 for teachings of Chai. Shattil further discloses wherein the neural network has three layers: the input layer, the output layer and a hidden layer between the input layer and the output layer. (as per “But each neuron in a hidden layer may receive multiple inputs, either from the input layer or from the outputs of neurons in an immediately preceding hidden layer. In general, each node may apply a function (e.g., an activation or transfer function) to its inputs to produce an output for that node. Nodes in hidden layers (e.g., learning layers) may apply the same function to their respective input(s) to produce their respective output(s).” in ¶7)
In this way, Shattil operates to update weights and/or biases to update in a learning process (¶6). Like Goupil, Shattil is concerned with aircraft systems. (¶164)
It would have been obvious for one of ordinary skill in the art before the effective filing date to have modified the speed estimation system of Goupil with the neural network system of Shattil to enable another standard means of estimating aircraft speed via a neural network. Such modification also allows the system to compute the gradient of the prediction output, error, or total loss with respect to a change in any of the ANN parameters. (¶152)
As per Claim 8, the combination of Goupil and Shattil teaches or suggests all limitations of Claim 6. Goupil fails to expressly disclose wherein the neural network is a feedforward neural network.
See Claim 6 for teachings of Chai. Shattil further discloses wherein the neural network is a feedforward neural network. (as per “a set of inputs is provided to an ANN, and a forward pass is performed, wherein outputs from one layer are fed forward to one or more other layers. An ANN, or neural net, is a (nodal) network of interconnected neurons, where each neuron represents a node in the network. Groups of neurons may be arranged in layers, with the outputs of one layer feeding forward to a next layer in a multilayer perception (MLP) arrangement. MLP may be understood to be a feedforward neural network model that maps a set of input data onto a set of output data. In some aspects, each neuron (or node) produces a single output that is fed forward to neurons in the layer immediately following it.” in ¶7)
In this way, Shattil operates to update weights and/or biases to update in a learning process (¶6). Like Goupil, Shattil is concerned with aircraft systems. (¶164)
It would have been obvious for one of ordinary skill in the art before the effective filing date to have modified the speed estimation system of Goupil with the neural network system of Shattil to enable another standard means of estimating aircraft speed via a neural network. Such modification also allows the system to compute the gradient of the prediction output, error, or total loss with respect to a change in any of the ANN parameters. (¶152)
As per Claim 9, the combination of Goupil and Shattil teaches or suggests all limitations of Claim 6.
Goupil fails to expressly disclose wherein the neural network is implemented by:
the pressure difference and the plurality of flight parameters corresponding to input variables are normalized,
each of the normalized input variables is multiplied by a synaptic weight and a bias is added to obtain a first weighting function for each of the neurons of the hidden layer,
each neuron of the hidden layer is activated by applying an activation function which is bounded to obtain a second weighting function,
the output layer linearly combines the weighting functions by multiplying the second weighting functions by the synaptic weights of the synapses connecting the neurons of the hidden layer to the neurons of the output layer and by adding a bias to obtain an output function,
a last operation comprises making the output function of the output layer homogeneous with a quantity to be estimated corresponding to the speed.
See Claim 6 for teachings of Chai. Shattil further discloses wherein the neural network is implemented by:
the pressure difference and the plurality of flight parameters corresponding to input variables are normalized, (as per “Data may be preprocessed 111 and input for training, testing, or online run. Data normalization may be employed, such as to normalize training and test sets.” in ¶166, as per “Batch norm 114 may be employed… batch norm parameters β[l] and γ[l] can be employed to compute the batch norm” in ¶173)
each of the normalized input variables is multiplied by a synaptic weight and a bias is added to obtain a first weighting function for each of the neurons of the hidden layer, (as per “The node combines its inputs, possibly with weights and/or possibly with a bias value, to produce a linear combination “z” value.” in ¶41, as per “At each node in each (l) hidden layer, the output from the previous layer is weighted and summed, and a bias may be added” in ¶171)
each neuron of the hidden layer is activated by applying an activation function which is bounded to obtain a second weighting function, (as per “The node may then perform a non-linear activation function on z to produce its output (e.g., activation).” in ¶41, as per “node 101.1 can comprise an activation function ƒ0L-1 that operates on the combined input z0L-1 and produces node output a0L-1” in ¶44)
the output layer linearly combines the weighting functions by multiplying the second weighting functions by the synaptic weights of the synapses connecting the neurons of the hidden layer to the neurons of the output layer and by adding a bias to obtain an output function, (as per “With reference to FIG. 1A, at the input to Layer L, which comprises one or more nodes, outputs a₀^(L–1), a₁^(L–1), … from Layer (L–1) may be weighted, and a bias may be added, to produce input vector zᴸ.” in ¶120)
a last operation comprises making the output function of the output layer homogeneous with a quantity to be estimated corresponding to the speed. (as per “The total error can be computed from the squared difference of each output value minus its corresponding target value, and these squared differences can be summed over all the outputs to produce the total error.” in ¶217)
In this way, Shattil operates to update weights and/or biases to update in a learning process (¶6). Like Goupil, Shattil is concerned with aircraft systems. (¶164)
It would have been obvious for one of ordinary skill in the art before the effective filing date to have modified the speed estimation system of Goupil with the neural network system of Shattil to enable another standard means of estimating aircraft speed via a neural network. Such modification also allows the system to compute the gradient of the prediction output, error, or total loss with respect to a change in any of the ANN parameters. (¶152)
As per Claim 11, the combination of Goupil and Shattil teaches or suggests all limitations of Claim 9. Goupil fails to expressly disclose wherein the activation function corresponds to a sigmoid of a type Θ(x) = 1
1
1
+
e
-
x
or to a “soft sign” function of type Θ(x) =
1
1
+
x
.
See Claim 9 for teachings of Shattil. Shattil further discloses wherein the activation function corresponds to a sigmoid of a type Θ(x) = 1
1
1
+
e
-
x
or to a “soft sign” function of type Θ(x) =
1
1
+
x
. (as per “such as depicted in FIGS. 3A and 3B, a PWL approximation may comprise dividing the nonlinear activation function into segments 301, which can include exploiting symmetry (e.g., in sigmoid, tanh, etc.) to reduce the number of computations.” in ¶88, as per “In one example, a sigmoid function y=f(x) can be implemented as follows. Since the sigmoid function has a symmetry point at (0, 0.5), only half of the x-y pairs need to be explicitly computed” in ¶92)
In this way, Shattil operates to update weights and/or biases to update in a learning process (¶6). Like Goupil, Shattil is concerned with aircraft systems. (¶164)
It would have been obvious for one of ordinary skill in the art before the effective filing date to have modified the speed estimation system of Goupil with the neural network system of Shattil to enable another standard means of estimating aircraft speed via a neural network. Such modification also allows the system to compute the gradient of the prediction output, error, or total loss with respect to a change in any of the ANN parameters. (¶152)
As per Claim 12, Goupil discloses a system for automatically estimating a speed of an aircraft during a flight of the aircraft (as per Abstract), comprising:
a determining step, implemented by a determining module in communication with a plurality of control surfaces of the aircraft, comprising determining, for each of the plurality of control surfaces of the aircraft, (as per “a data generating unit 2 configured to determine at least one aerodynamic hinge moment of at least one control surface 3 of the aircraft;” in ¶62-¶64, as per “This aerodynamic hinge moment illustrates the aerodynamic forces acting on the control surface of the aircraft in the course of the flight” in ¶19)
at least one quantity which is representative of a force exerted on the corresponding one of the plurality of control surfaces of the aircraft (as per “measuring at least one pressure difference between two hydraulic chambers of at least one actuator intended to generate the deflection of the control surface; and computing the aerodynamic hinge moment, using at least the pressure difference” in Claim 2), each of the plurality of control surfaces comprising an actuator configured to deflect a corresponding one of the plurality of control surfaces of the aircraft, (as per “the device 1 comprises a set 11 of units 2 for determining aerodynamic hinge moments of a plurality of different control surfaces 3 of the aircraft, and the central unit 5 computes the Mach number” in ¶87, as per “the deflection of this control surface 3 is generated by one or more actuators 12 which can be of any type, in particular of hydraulic type, of electro-hydrostatic type or of electromechanical type” in ¶79)
the at least one quantity which is representative of a force each corresponding to a pressure difference measured between two hydraulic chambers of the actuator to deflect the corresponding one of the plurality of control surfaces of the aircraft; (as per “measuring at least one pressure difference between two hydraulic chambers of at least one actuator intended to generate the deflection of the control surface; and computing the aerodynamic hinge moment, with the aid at least of this pressure difference.” in ¶20)
a calculating step, implemented by a calculating module, comprising calculating at least one speed of the aircraft at least based on the at least one quantity corresponding to each of the plurality of control surfaces determined by the determining module; (as per “computing a Mach number M representing a first speed of the aircraft,” in ¶6-¶18, as per “the speed of the aircraft is estimated automatically, in real time, on the basis of the value of at least one aerodynamic hinge moment of at least one control surface of the aircraft, without using any measurement of the total air pressure, thereby making it possible to dispense with the associated sensors, generally redundant on civil airplanes.” in ¶18)
a transmitting step, implemented by a transmitting module, comprising transmitting the at least one speed of the aircraft calculated in the calculating step to a user device; (as per “at least one data transmission unit configured to provide the first speed to at least one user system,” in ¶38)
a collecting step, implemented by a collecting module, comprising collecting a plurality of flight parameters from the aircraft, (as per “b) determining a plurality of data and at least the following data: a static pressure external to the aircraft; and a model of hinge moment coefficient;” in ¶6-¶18)
the at least one speed of the aircraft calculated by the calculating step based on the pressure difference and the plurality of flight parameters using a function in which the at least one speed is a function of the pressure difference, (as per “the following expression: Mₐ = ½ [γ · Pₛ · M² · Vδₚ · Cₕ]… Ma is the aerodynamic hinge moment; γ is an adiabatic coefficient of the air; Ps is an external static pressure; Vδ p represents a volume of the control surface; and Ch is the model of aerodynamic hinge moment coefficient of the control surface, dependent on the Mach number M;” in ¶6-¶18)
the function having the following expression:
VCAS =f(θ),
in which VCAS corresponds to the at least one speed to be determined, (as per “the method comprises an additional step, posterior to step c) and consisting in or comprising computing a conventional speed (or corrected speed) CAS, representing a second speed of the aircraft, on the basis of the Mach number computed in step c) and of additional data.” in ¶36)
θ corresponds to a parameter vector comprising the pressure difference determined for each of the plurality of control surfaces of the aircraft by the determining step and the plurality of flight parameters which are measured by the collecting step; (as per “the following expression: Mₐ = ½ [γ · Pₛ · M² · Vδₚ · Cₕ]… Ma is the aerodynamic hinge moment; γ is an adiabatic coefficient of the air; Ps is an external static pressure; Vδ p represents a volume of the control surface; and Ch is the model of aerodynamic hinge moment coefficient of the control surface, dependent on the Mach number M;” in ¶6-¶18)
Goupil fails to expressly disclose:
the at least one speed being calculated based on a neural network in which the at least one speed to be calculated corresponds to an output layer of the neural network and
in which the pressure difference and the plurality of flight parameters correspond to an input layer of the neural network,
the neural network comprising fixed synaptic weights which are determined off-line through training on data sets which characterize a correspondence between the vector parameter and the at least one speed determined for a plurality of flights of the aircraft.
Shattil discloses of an artificial neural network (as per Abstract), comprising:
the at least one speed being calculated based on a neural network in which the at least one speed to be calculated corresponds to an output layer of the neural network (as per “layer l might be the input layer, and z0 l can be the ANN input value, x0. In one example, layer l might be the output layer, and a0l can be the ANN output value y0..” in ¶34,)
in which the pressure difference and the plurality of flight parameters correspond to an input layer of the neural network, (as per “The disclosed ANNs can include any type of sensor(s) for providing input data to the ANNs. Sensors can comprise … and/or any type of situational awareness sensors and systems (e.g., accelerometer, inclinometer, inertial sensor, orientation sensor, speedometer, altimeter, primary flight displays (PFDs), flight management systems (FMS), collision avoidance systems (CAS), air traffic control (ATC) systems, navigation systems (e.g., pointing, navigation, and timing receivers), gyroscope, tachometer, oil pressure gauge, fuel gauge, etc.).” in ¶164, as per “data inputs to the ANN can comprise measurements of physical signals, such as Physical-Layer signals used for data communications. The input data may include a signal received in a wireless communication network and/or a signal to be transmitted in the wireless communication network.” in ¶22)
the neural network comprising fixed synaptic weights which are determined off-line through training on data sets which characterize a correspondence between the vector parameter and the at least one speed determined for a plurality of flights of the aircraft. (as per “The weights are then trained (optimized) so that for a given training vector input, the neural network produces an output close to a desired (predetermined) training vector output.” in ¶8, as per “An error for each output neuron, or output node, is then calculated based on the actual neuron output and a target training output for that neuron. Then backpropagation may be performed through the neural network (in a direction from the output layer back to the input layer) to update the weights based on how much effect each weight has on the overall error, so that the output of the neural network moves closer to the desired training output. This cycle can be repeated until the actual output of the neural network is within an acceptable error range of the desired training output.” in ¶8, as per “The error and/or total cost may be used to update parameters in the ANN, such as to reduce the error of the ANN's output, or prediction, such as may be measured from a labeled training data set” in ¶11)
In this way, Shattil operates to update weights and/or biases to update in a learning process (¶6). Like Goupil, Shattil is concerned with aircraft systems. (¶164)
It would have been obvious for one of ordinary skill in the art before the effective filing date to have modified the speed estimation system of Goupil with the neural network system of Shattil to enable another standard means of estimating aircraft speed via a neural network. Such modification also allows the system to compute the gradient of the prediction output, error, or total loss with respect to a change in any of the ANN parameters. (¶152)
As per Claim 13, the combination of Goupil and Shattil teaches or suggests all limitations of Claim 1. Goupil further discloses an aircraft comprising a system for automatically estimating a speed of an aircraft during a flight of the aircraft. (as per Claim 14, as per “The present disclosure also relates to an aircraft, in particular a transport airplane, which is provided with a device such as that specified hereinabove.” in ¶54)
As per Claim 14, the combination of Goupil and Shattil teaches or suggests all limitations of Claim 1. Goupil fails to expressly disclose wherein the neural network is configured to perform a non-linear regression to estimate the at least one speed of the aircraft based on a relationship between the pressure difference and the flight parameters.
See Claim 1 for teachings of Shattil. Shattil further disclose wherein the neural network is configured to perform a non-linear regression (as per “A trained feedforward neural network can be regarded as a nonlinear mapping from the input space to the output space. “ in ¶4, as per “Nodes in hidden layers (e.g., learning layers) may apply the same function to their respective input(s) to produce their respective output(s).” in ¶7) to estimate the at least one speed of the aircraft based on a relationship between the pressure difference and the flight parameters. (as per “data inputs to the ANN can comprise measurements of physical signals, such as Physical-Layer signals used for data communications” in ¶22, as per “The disclosed ANNs can include any type of sensor(s) for providing input data to the ANNs. Sensors can comprise … and/or any type of situational awareness sensors and systems (e.g., accelerometer, inclinometer, inertial sensor, orientation sensor, speedometer, altimeter, primary flight displays (PFDs), flight management systems (FMS), collision avoidance systems (CAS), air traffic control (ATC) systems, navigation systems (e.g., pointing, navigation, and timing receivers), gyroscope, tachometer, oil pressure gauge, fuel gauge, etc.).” in ¶164, as per “layer l might be the input layer, and z0 l can be the ANN input value, x0. In one example, layer l might be the output layer, and a0l can be the ANN output value y0..” in ¶34,)
In this way, Shattil operates to update weights and/or biases to update in a learning process (¶6). Like Goupil, Shattil is concerned with aircraft systems. (¶164)
It would have been obvious for one of ordinary skill in the art before the effective filing date to have modified the speed estimation system of Goupil with the neural network system of Shattil to enable another standard means of estimating aircraft speed via a neural network. Such modification also allows the system to compute the gradient of the prediction output, error, or total loss with respect to a change in any of the ANN parameters. (¶152)
As per Claim 15, the combination of Goupil and Shattil teaches or suggests all limitations of Claim 1. Goupil further discloses wherein each of the plurality of control surfaces of the aircraft comprises one of an aileron or an elevator of a rudder of the aircraft. (as per “Thus, although not exclusively, this control surface may in particular correspond to any type of control surface: aileron, elevator, spoiler, rudder, etc. of the aircraft” in ¶78, as per “the device 1 comprises a set 11 of units 2 for determining aerodynamic hinge moments of a plurality of different control surfaces 3 of the aircraft, and the central unit 5 computes the Mach number” in ¶87)
Claims 3 and 4 are rejected under 35 U.S.C. 103 as being unpatentable over Goupil (US Pub. No. 20160107762) in view of Shattil (US Pub. No. 20200364545) in further view of Hickman (US Pub No. 20110202291).
As per Claim 3, the combination of Goupil and Shattil teaches or suggests all limitations of Claim 1. Goupil and Shattil fail to expressly disclose wherein the determining module comprises a filtering submodule configured to filter one or more individual pressure differences collected.
Hickman discloses of a dynamic pressure estimation system (as per Abstract), wherein the determining module comprises a filtering submodule configured to filter one or more individual pressure differences collected. (as per “estimating dynamic pressure applied to an aircraft having a flight control surface actuator coupled to a flight control surface includes measuring a load on the flight control surface actuator. An estimate of the dynamic pressure is calculated from the measured load.” in ¶5, as per “After suitable signal processing 202, a first summing function 204 sums the actuator load signals from the load sensors 122 associated with the left aileron 102 to determine the left aileron load, and a second summing function 206 sums the actuator load signals from the load sensors 122 associated with the right aileron 104 to determine the right aileron load. An averaging function 208 then averages the left and right aileron loads and, after suitable filtering by a low-pass filter 210, the average aileron load is supplied to a conversion function 212... The conversion function 212 calculates the estimate of aircraft dynamic pressure (q) from the average aileron load.” in ¶32-¶33)
In this way, Hickman operates to calculate an estimate of aircraft dynamic pressure from flight control surface actuator load sensors. (¶1) Like Goupil and Shattil, Hickman is concerned with aircraft. (Abstract)
It would have been obvious for one of ordinary skill in the art before the effective filing date to have modified the speed estimation system of Goupil and the neural network system of Shattil with the dynamic pressure estimation of Hickman to enable another standard means of processing the pressure differences collected from the control surfaces of an aircraft.
As per Claim 4, the combination of Goupil and Shattil teaches or suggests all limitations of Claim 2. Goupil and Shattil fail to expressly disclose wherein the determining module comprises a determining submodule configured to determine an overall pressure difference by calculating a mean, or a median, or a weighted mean of the individual pressure difference.
Hickman discloses of a dynamic pressure estimation system (as per Abstract), wherein the determining module comprises a determining submodule configured to determine an overall pressure difference by calculating a mean, or a median, or a weighted mean of the individual pressure difference. (as per “estimating dynamic pressure applied to an aircraft having a flight control surface actuator coupled to a flight control surface includes measuring a load on the flight control surface actuator. An estimate of the dynamic pressure is calculated from the measured load.” in ¶5, as per “After suitable signal processing 202, a first summing function 204 sums the actuator load signals from the load sensors 122 associated with the left aileron 102 to determine the left aileron load, and a second summing function 206 sums the actuator load signals from the load sensors 122 associated with the right aileron 104 to determine the right aileron load. An averaging function 208 then averages the left and right aileron loads and, after suitable filtering by a low-pass filter 210, the average aileron load is supplied to a conversion function 212... The conversion function 212 calculates the estimate of aircraft dynamic pressure (q) from the average aileron load.” in ¶32-¶33)
In this way, Hickman operates to calculate an estimate of aircraft dynamic pressure from flight control surface actuator load sensors. (¶1) Like Goupil and Shattil, Hickman is concerned with aircraft. (Abstract)
It would have been obvious for one of ordinary skill in the art before the effective filing date to have modified the speed estimation system of Goupil and the neural network system of Shattil with the dynamic pressure estimation of Hickman to enable another standard means of processing the pressure differences collected from the control surfaces of an aircraft.
Claim 7 is rejected under 35 U.S.C. 103 as being unpatentable over Goupil (US Pub. No. 20160107762) in view of Shattil (US Pub. No. 20200364545) in further view of Hao (CN Pub No. 112597700).
As per Claim 7, the combination of Goupil and Shattil teaches or suggests all limitations of Claim 6. Goupil and Shattil fail to expressly disclose wherein the hidden layer comprises a maximum number of neurons which is less than 20.
Hao discloses of a trajectory simulation based on a neural network (as per Abstract), wherein the hidden layer comprises a maximum number of neurons which is less than 20. (as per “the hidden layer node number is 10” in pg. 2, ¶4)
In this way, Hao operates to improve speed and precision of neural networks. (Abstract) Like Goupil and Shattil, Hao is concerned with neural networks. (Abstract)
It would have been obvious for one of ordinary skill in the art before the effective filing date to have modified the speed estimation system of Goupil and the neural network system of Shattil with the aircraft simulation based neural network of Hao to enable another standard means of modifying the number of neurons in the hidden layer.
Claim 10 is rejected under 35 U.S.C. 103 as being unpatentable over Goupil (US Pub. No. 20160107762) in view of Shattil (US Pub. No. 20200364545) in further view of Brownlee (NPL Title: How to use Data Scaling Improve Deep Learning Model Stability and Performance, Year 2019).
As per Claim 10, the combination of Goupil and Shattil teaches or suggests all limitations of Claim 1. Goupil and Shattil fail to expressly disclose wherein the input variables are normalized by:
the input variables are lower-bounded by zero,
the input variables are then normalized between −1 and +1,
the output function of the output layer being made homogeneous by:
the output function is lower-bounded at zero,
the output function is then normalized with respect to maximum and minimum values of the speed which is observed during training.
Brownlee discloses of data scaling to improve deep learning model stability and performance (as per Title), wherein the input variables are normalized by:
the input variables are lower-bounded by zero, (as per “Normalization is a rescaling of the data from the original range so that all values are within the range of 0 and 1.” in pg. 2, Data Scaling Methods)
the input variables are then normalized between −1 and +1, (as per “scaler = MinMaxScaler(feature_range=(-1,1)” in pg. 2, Data Scaling Methods)
the output function of the output layer being made homogeneous by:
the output function is lower-bounded at zero, (as per “If your output activation function has a range of [0,1], then obviously you must ensure that the target values lie within that range” in pg. 2, Scaling Output Variables, as per “Normalization is a rescaling of the data from the original range so that all values are within the range of 0 and 1.” in pg. 2, Data Scaling Methods)
the output function is then normalized with respect to maximum and minimum v