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 .
DETAILED ACTION
Response to Amendment
The amendment filed 10/20/2025, has been entered. Claims 1-14 are pending in the application. Applicant’s amendments to the claims have overcome each and every 35 U.S.C. 101 Rejection previously set forth in the Final Action mailed 4/22/2025.
Response to Arguments
Applicant’s arguments with respect to claim(s) 1-11 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
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.
Claims 1-14 are rejected under 35 U.S.C. 103 as being unpatentable over Schat (U.S. Patent Application Pub. No. 2019/0331555) in view of Cheng (“A Neural Network Approach to Ordinal Regression”, 2007), and further in view of Ohlarik, et al., hereinafter Ohlarik (U.S. Patent Application Pub. No. 2021/0312811), Martin, et al., hereinafter Martin (WIPO Patent App. Pub. No. WO 2020/176914 A1), and Mahdi, et al., hereinafter Mahdi (“Modeling Diesel Oxidation Catalyst Upstream and Downstream Exhaust Gas Temperatures Using LSTM RNN”, 2018).
Regarding Claim 1, Schat teaches: A method for manipulation detection of a technical device (Schat, Abstract and Para. 0001 – a method for “detecting defeat devices in an engine control unit”; where defeat devices are used to manipulate results of tests), the method comprising the following steps:
providing an input vector including one or multiple system variables (Schat, Para. 0018, 0032-0033, and 0041 – “input data” that is input for a “machine-learning algorithm”, where the input data is indicative of “different input conditions” for testing; where the testing input conditions can be related to “engine temperatures”, speed changes, “air temperature”, “reduced cooling capacity of the cooling system”, fuel consumption, etc.; where the input data constitutes an input vector within a coded algorithm) and including at least one control variable for an intervention in the technical device, for successive time steps (Schat, Para. 0032-0033 and 0036-0042 – where input data is used to create a “test environment”, where input conditions include constants, or controls, which are set in the test, such as “a time constant within a certain time range”, a constant speed for several minutes, an air temperature staying constant, etc.; where the test is conducted over a period of time due to the behavior of defeat devices being “time dependent”);
using a data-based manipulation detection model (Schat, Para. 0025-0026 – “a machine-learning method” which accumulates knowledge of environmental test results to anticipate behavior of defeat devices) to generate a corresponding output vector where the machine learning includes an estimation methods, such that the machine learning method will determine, or output, which “data positions” and “time slots” have the highest likelihood “of changing the defeat device's behavior”, such that the likelihood is the output; where the output constitutes an output vector within a coded algorithm and where the machine learning is based on “input data”),
providing an actual monitored variable based on an at least one measured value in the successive time steps (Schat, Para. 0025-0026, 0032, and 0042 – where the machine learning method determines “data positions”, or an actual monitored variable, where the machine learning method is based on collected key-data, or measured values; where the data behavior is time dependent and the output includes “time slots”);
creating a where the machine learning method estimates a highest likelihood of a defeat device changing behavior; where the likelihood is the classification and the data used is time dependent); and
detecting a manipulation as a function of the measurement classification vector and a first and a second comparison vector for one or multiple of the time steps of a time window (Schat, Para. 0042 – where a comparison of key-data and ECU data “will indicate the presence of a defeat device, if there is a difference in the comparison (e.g., an anomaly)”), the first and the second comparison vector being determined by rounding element values of the output vector based on a first manipulation threshold value and a second manipulation threshold value, which is different from the first manipulation threshold value, as rounding limits (Schat, Para. 0042 – where “key-data is considered to be “similar” when changes are due merely to uncontrollable random variations or the input data and internal states. In one embodiment, a level of these random variations is predetermined by adding a threshold value below which a pair of key-data is considered to be similar.”);
providing a controller that controls, based on the control variable, a physical action of the technical device (Schat, Para. 0001, 0015, 0032-0033 and 0036-0042 – where a “urea-based exhaust after-treatment system” is activated or deactivated based on when a “defeat device” detects an “environmental test condition”; where the environmental test condition set by input data used to create a “test environment”, where input conditions include constants, or controls, which are set in the test).
While Schat teaches using a data-based manipulation detection model to generate a corresponding output vector, the data-based manipulation detection model being configured to output an output vector, for the input vector, Schat does not fully teach outputting an output vector as a classification vector, which indicates a classification of a monitored variable in value ranges, for the input vector. Additionally, while Schat teaches creating an output vector from the actual monitored variable for each time step, Schat does not teach creating a measurement classification vector from the actual monitored variable for each time step. Schat does not teach wherein the data-based manipulation model includes a neural network that contains recurrent components that permit the data-based manipulation detection model to take into consideration a dynamic behavior of the technical device, wherein the neural network includes a fully connected output layer that outputs the output vector.
However, Cheng teaches outputting an output vector as a classification vector, which indicates a classification of a monitored variable in value ranges, for the input vector (Cheng, 2. Method – where a machine learning model receives an input vector and outputs a “probability distribution vector” and considers the vector as “a cumulative probability distribution on categories (1, ..., k, ..., K)”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Schat to include outputting an output vector as a classification vector, which indicates a classification of a monitored variable in value ranges, for the input vector, as taught by Cheng, in order to allow the machine learning model to train on larger data sets and make faster predictions once trained on the categorized data (Cheng, Abstract and 4. Discussion and Future Work).
Schat in view of Cheng does not teach creating a measurement classification vector from the actual monitored variable for each time step, and wherein the data-based manipulation model includes a neural network that contains recurrent components that permit the data-based manipulation detection model to take into consideration a dynamic behavior of the technical device, and wherein the data-based manipulation model includes a neural network that contains recurrent components that permit the data-based manipulation detection model to take into consideration a dynamic behavior of the technical device.
However, Ohlarik teaches creating a measurement classification vector from the actual monitored variable for each time step (Ohlarik, Para. 0067-0069 – where a machine learning model takes in vectors which represent changes in a measurement, in this case location, for “consecutive time intervals”, and generates an output vector which represents a likelihood of a collision “occurring during a particular time interval”, where the classes of the vector represent specific time steps such as 0.1 seconds, 0.2 seconds, etc.; where the likelihood is the classification).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the method of Schat in view of Cheng to include creating a measurement classification vector from the actual monitored variable for each time step, as taught by Ohlarik, in order to measure the changes within the technical device over time steps to determine if there is manipulation occurring over time.
Schat in view of Cheng and Ohlarik does not teach wherein the data-based manipulation model includes a neural network that contains recurrent components that permit the data-based manipulation detection model to take into consideration a dynamic behavior of the technical device, wherein the neural network includes a fully connected output layer that outputs the output vector.
However, Martin teaches wherein the data-based manipulation model includes a neural network (Martin, Para. 0003-0004 and 0023 – using “artificial neural networks” to create a “mathematical model” representative of a “exhaust gas aftertreatment component of an exhaust gas aftertreatment system”) that contains recurrent components that permit the data-based manipulation detection model to take into consideration a dynamic behavior of the technical device (Martin, Para. 0023, 0032-0034, 0072 and 0094 – where the neural network used for creating a model can be a “recurrent neural network”, where the neural network is trained and adapted simultaneously, such that it is dynamic), wherein the neural network includes a fully connected output layer (Martin, Fig. 1 and Para. 0109 – where Fig. 1 shows the “neural network” having an “output layer”, where as shown on Fig. 1, is fully connected to the nodes of the other layers of the neural network).
PNG
media_image1.png
756
777
media_image1.png
Greyscale
Martin, Fig. 1
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the method including the above limitations of Schat in view of Cheng and Ohlarik to include wherein the data-based manipulation model includes a neural network that contains recurrent components that permit the data-based manipulation detection model to take into consideration a dynamic behavior of the technical device, as taught by Martin, in order to reduce the computational effort and runtime of the model (Martin, Para. 0037).
Schat in view of Cheng, Ohlarik, and Martin does not explicitly teach wherein the neural network includes an output layer that outputs the output vector.
However, Mahdi teaches wherein the neural network includes an output layer that outputs the output vector (Mahdi, Section I – where it is known in the art that recurrent neural networks (RNN) have the ability to “build on earlier types of networks with fixed-size input and output vectors”, where Mahdi proposes a “many-to-many” network structure with multiple layers which provides “output vectors” that predict “DOC upstream and downstream exhaust gas temperatures”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the method including the above limitations of Schat in view of Cheng, Ohlarik, and Martin, to include wherein the neural network includes an output layer that outputs the output vector, as taught by Mahdi, in order to provide a model which produces many outputs which reflect the state of the technical device.
In regards to Claim 2, Schat in view of Cheng, Ohlarik, Martin, and Mahdi teaches the method of Claim 1, and Schat further teaches wherein the technical device is an exhaust gas aftertreatment device in a motor vehicle (Schat, Para. 0001 and 0015 – where in one example, a defeat device is used to deactivate “a urea-based exhaust after-treatment system for a diesel vehicle”, such that it manipulates the after-treatment system, outside testing environments, to “provide better performance”).
In regards to Claim 3, Schat in view of Cheng, Ohlarik, Martin, and Mahdi teaches the method of Claim 1, and Schat teaches the output vector and the monitored variable but Schat does not teach wherein the output vector includes a nominal coding which indicates for the monitored variable in which value ranges the monitored variable lies, the value ranges being classified by a number of classes, the value ranges of the monitored variable including ascending index values k of the output vector each being indicated by corresponding ascending/descending classification threshold values S1, S2, S3, …, SK-1, the threshold values indicating with their value whether the monitored variable is expected to be less or greater than the classification threshold value corresponding to the index value of the element of the output vector.
However, Cheng teaches wherein the output vector includes a nominal coding which indicates for the monitored variable in which value ranges the monitored variable lies, the value ranges being classified by a number of classes, the value ranges of the monitored variable including ascending index values k of the output vector each being indicated by corresponding ascending/descending classification threshold values S1, S2, S3, …, SK-1, (Cheng, 2. Method – “an ordinal regression dataset of n data points (x,y)” where x is an input vector and y is its category [organized such that the input vector is sorted into lower order categories of an ascending index “(1, 2, ..., k − 1)”] and the goal is to obtain a probability, or output, vector) the threshold values indicating with their value whether the monitored variable is expected to be less or greater than the classification threshold value corresponding to the index value of the element of the output vector (Cheng, 2. Method – where the values are sorted into the categories based on how close to a threshold value they are; where k is the index corresponding to the categories).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the method including the above limitations of Schat in view of Cheng, Ohlarik, Martin, and Mahdi to further include wherein the output vector includes a nominal coding which indicates for the monitored variable in which value ranges the monitored variable lies, the value ranges being classified by a number of classes, the value ranges of the monitored variable including ascending index values k of the output vector each being indicated by corresponding ascending/descending classification threshold values S1, S2, S3, …, SK-1, the threshold values indicating with their value whether the monitored variable is expected to be less or greater than the classification threshold value corresponding to the index value of the element of the output vector, as taught by Cheng, in order to implement a method to allow the machine learning model to train on larger data sets, such that the model can be more accurate, and make faster predictions once trained (Cheng, Abstract and 4. Discussion and Future Work).
In regards to Claim 4, Schat in view of Cheng, Ohlarik, Martin, and Mahdi teaches the method of Claim 3, and Schat in view of Cheng, Ohlarik, Martin, and Mahdi teaches wherein the measurement classification vector including a nominal coding is created using the value of the actual monitored variable (Schat, Para. 0025-0026 and 0048 – where the machine learning method outputs a probability of change caused by a defeat device, and where a time variant data set is “categorize[d]” as either “a constant value set and a changing value set”, or nominally coded; Ohlarik, Para. 0067-0069 – where a machine learning model takes in vectors which represent changes in a measurement, in this case location, for “consecutive time intervals”, and generates an output vector which represents a likelihood of a collision “occurring during a particular time interval”; where the likelihood is the classification), but Schat does not teach the elements of the measurement classification vector having a first value when the actual monitored variable is expected to be less or greater than the classification threshold value corresponding to the index value of the element of the output vector and having a second value when the actual monitored variable is expected to be greater or less than the classification threshold value corresponding to the index value of the element of the output vector.
However, Cheng teaches the elements of the measurement classification vector having a first value when the actual monitored variable is expected to be less or greater than the classification threshold value corresponding to the index value of the element of the output vector and having a second value when the actual monitored variable is expected to be greater or less than the classification threshold value corresponding to the index value of the element of the output vector (Cheng, 2. Method – where a probability distribution vector o = (o1, o2, ...ok, ...oK), or classification vector, maps the input vectors, where “oi (i ≤ k) is close to 1 and oi (i ≥ k) is close to 0”, where k is the index and 1 and 0 are the threshold values corresponding to the index k).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the method including the above limitations of Schat in view of Cheng, Ohlarik, Martin, and Mahdi to further include the elements of the measurement classification vector having a first value when the actual monitored variable is expected to be less or greater than the classification threshold value corresponding to the index value of the element of the output vector and having a second value when the actual monitored variable is expected to be greater or less than the classification threshold value corresponding to the index value of the element of the output vector, as taught by Cheng, in order to implement a method to allow the machine learning model to train on larger data sets, such that the model can be more accurate, and make faster predictions once trained (Cheng, Abstract and 4. Discussion and Future Work).
In regards to Claim 5, Schat in view of Cheng, Ohlarik, Martin, and Mahdi teaches the method of Claim 4, and Schat in view of Cheng, Ohlarik, Martin, and Mahdi further teaches wherein to determine the first comparison vector, the elements of the output vector are rounded to the first value based on exceeding the first manipulation threshold value as a rounding limit and to a second value based on not reaching the first manipulation threshold value as a rounding limit, to determine the second comparison vector (Schat, Para. 0042 – where “key-data is considered to be “similar” when changes are due merely to uncontrollable random variations or the input data and internal states. In one embodiment, a level of these random variations is predetermined by adding a threshold value below which a pair of key-data is considered to be similar”; where the threshold value is the manipulation threshold value), the elements of the output vector being rounded to the first value based on exceeding the second manipulation threshold value as a rounding limit and being rounded to the second value based on not reaching the second manipulation threshold value as a rounding limit (Cheng, 2. Method – where a probability vector, or output vector, is mapped based on whether a value is close to 1 or 0, such that the values are rounded to the threshold values), the manipulation being recognized as a function of a difference between the number of the element values of the first comparison vector having the first value and the number of the element values of the measurement classification vector having the first value and as a function of a difference between the number of the element values of the measurement classification vector having the first value and the number of the element values of the second comparison vector having the first value (Schat, Para. 0025-0026, 0042 and 0048 – where the first and second key-datas, or comparison vectors, are compared and an anomaly is detected based on “is determined by a difference between the first comparison and the second comparison”; where a likelihood “of changing the defeat device's behavior”, or classification, is determined for the data sets).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the method including the previous limitations, as taught by Schat in view of Cheng, Ohlarik, Martin, and Mahdi, to further include the elements of the output vector being rounded to the first value based on exceeding the second manipulation threshold value as a rounding limit and being rounded to the second value based on not reaching the second manipulation threshold value as a rounding limit, , as taught by Cheng, in order to implement a method to allow the machine learning model to train on larger data sets, such that the model can be more accurate, and make faster predictions once trained (Cheng, Abstract and 4. Discussion and Future Work).
In regards to Claim 6, Schat in view of Cheng, Ohlarik, Martin, and Mahdi teaches the method of Claim 1, and Schat further teaches wherein for each time window, a manipulation signal is generated, a manipulation being recognized as a function of a portion of the manipulation signals indicating a manipulation for multiple time windows of an evaluation time period (Schat, Para. 0018, 0032, and 0042 – where a display is configured to display a status flag “upon detection of an anomaly”, such that a signal is sent to the display by communication through a bus; where an anomaly “will indicate the presence of a defeat device, if there is a difference in [a] comparison”; where the behavior measured is time dependent).
In regards to Claim 7, Schat in view of Cheng, Ohlarik, Martin, and Mahdi teaches the method of Claim 1, and Schat further teaches wherein the technical device includes an exhaust gas aftertreatment device (Schat, Para. 0001 and 0015 – where in one example, a defeat device is used to deactivate “a urea-based exhaust after-treatment system for a diesel vehicle”, such that it manipulates the after-treatment system, outside testing environments, to “provide better performance”), the input vector including as the control variable a control variable for a urea injection system (Schat, Para. 0015 and 0036-0042 – where the input data includes constants that are set in a test; where a manipulating defeat device is used in a urea-based exhaust after-treatment system).
In regards to Claim 8, Schat in view of Cheng, Ohlarik, Martin, and Mahdi teaches the method of Claim 1, and Schat further teaches wherein a recognized manipulation is signaled or the technical device is operated as a function of the recognized manipulation (Schat, Para. 0018 and 0042– where a display “is configured to display a status flag upon detection of an anomaly”, or signals through the status flag, and an “anomaly” is a difference between a stored key-data, measured in a test environment, and an ECU data, measured in a non-test environment).
Regarding Claim 9, Schat teaches: A device for manipulation detection of a technical device (Schat, Abstract, Para. 0001 and 0025 – a method for “detecting defeat devices in an engine control unit” implemented by a “Environmental Testing Device”; where defeat devices are used to manipulate results of tests), the device being configured to:
provide an input vector including one or multiple system variables (Schat, Para. 0018, 0032-0033, and 0041 – “input data” that is input for a “machine-learning algorithm”, where the input data is indicative of “different input conditions” for testing; where the testing input conditions can be related to “engine temperatures”, speed changes, “air temperature”, “reduced cooling capacity of the cooling system”, fuel consumption, etc.; where the input data constitutes an input vector within a coded algorithm) and including at least one control variable for an intervention in the technical device, for successive time steps (Schat, Para. 0032-0033 and 0036-0042 – where input data is used to create a “test environment”, where input conditions include constants, or controls, which are set in the test, such as “a time constant within a certain time range”, a constant speed for several minutes, an air temperature staying constant, etc.; where the test is conducted over a period of time due to the behavior of defeat devices being “time dependent”);
use a data-based manipulation detection model (Schat, Para. 0025-0026 – “a machine-learning method” which accumulates knowledge of environmental test results to anticipate behavior of defeat devices) to generate a corresponding output vector where the machine learning includes an estimation methods, such that the machine learning method will determine, or output, which “data positions” and “time slots” have the highest likelihood “of changing the defeat device's behavior”, such that the likelihood is the output; where the output constitutes an output vector within a coded algorithm and where the machine learning is based on “input data”),
provide an actual monitored variable based on an at least one measured value in the successive time steps (Schat, Para. 0025-0026, 0032, and 0042 – where the machine learning method determines “data positions”, or an actual monitored variable, where the machine learning method is based on collected key-data, or measured values; where the data behavior is time dependent and the output includes “time slots”);
create a where the machine learning method estimates a highest likelihood of a defeat device changing behavior; where the likelihood is the classification and the data used is time dependent);
recognize a manipulation as a function of the measurement classification vector and a first and a second comparison vector for one or multiple of the time steps of a time window (Schat, Para. 0042 – where a comparison of key-data and ECU data “will indicate the presence of a defeat device, if there is a difference in the comparison (e.g., an anomaly)”), the first and the second comparison vector being determined by rounding element values of the output vector based on a first manipulation threshold value and a second manipulation threshold value, which is different from the first, as rounding limits (Schat, Para. 0042 – where “key-data is considered to be “similar” when changes are due merely to uncontrollable random variations or the input data and internal states. In one embodiment, a level of these random variations is predetermined by adding a threshold value below which a pair of key-data is considered to be similar.”),
providing a controller controls, based on the control variable, a physical action of the technical device (Schat, Para. 0001, 0015, 0032-0033 and 0036-0042 – where a “urea-based exhaust after-treatment system” is activated or deactivated based on when a “defeat device” detects an “environmental test condition”; where the environmental test condition set by input data used to create a “test environment”, where input conditions include constants, or controls, which are set in the test).
While Schat teaches using a data-based manipulation detection model to generate a corresponding output vector, the data-based manipulation detection model being configured to output an output vector, for the input vector, Schat does not fully teach outputting an output vector as a classification vector, which indicates a classification of a monitored variable in value ranges, for the input vector. Additionally, while Schat teaches creating an output vector from the actual monitored variable for each time step, Schat does not teach creating a measurement classification vector from the actual monitored variable for each time step, and Schat does not teach wherein the data-based manipulation model includes a neural network that contains recurrent components that permit the data-based manipulation detection model to take into consideration a dynamic behavior of the technical device, wherein the neural network includes a fully connected output layer that outputs the output vector.
However, Cheng teaches outputting an output vector as a classification vector, which indicates a classification of a monitored variable in value ranges, for the input vector (Cheng, 2. Method – where a machine learning model receives an input vector and outputs a “probability distribution vector” and considers the vector as “a cumulative probability distribution on categories (1, ..., k, ..., K)”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Schat to include outputting an output vector as a classification vector, which indicates a classification of a monitored variable in value ranges, for the input vector, as taught by Cheng, in order to allow the machine learning model to train on larger data sets and make faster predictions once trained on the categorized data (Cheng, Abstract and 4. Discussion and Future Work).
Schat in view of Cheng does not teach creating a measurement classification vector from the actual monitored variable for each time step and wherein the data-based manipulation model includes a neural network that contains recurrent components that permit the data-based manipulation detection model to take into consideration a dynamic behavior of the technical device, wherein the neural network includes a fully connected output layer that outputs the output vector.
However, Ohlarik teaches creating a measurement classification vector from the actual monitored variable for each time step (Ohlarik, Para. 0067-0069 – where a machine learning model takes in vectors which represent changes in a measurement, in this case location, for “consecutive time intervals”, and generates an output vector which represents a likelihood of a collision “occurring during a particular time interval”, where the classes of the vector represent specific time steps such as 0.1 seconds, 0.2 seconds, etc.; where the likelihood is the classification).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the device of Schat in view of Cheng to include creating a measurement classification vector from the actual monitored variable for each time step, as taught by Ohlarik, in order to measure the changes within the technical device over time steps to determine if there is manipulation occurring over time.
Schat in view of Cheng and Ohlarik does not teach wherein the data-based manipulation model includes a neural network that contains recurrent components that permit the data-based manipulation detection model to take into consideration a dynamic behavior of the technical device.
However, Martin teaches wherein the data-based manipulation model includes a neural network (Martin, Para. 0003-0004 and 0023 – using “artificial neural networks” to create a “mathematical model” representative of a “exhaust gas aftertreatment component of an exhaust gas aftertreatment system”) that contains recurrent components that permit the data-based manipulation detection model to take into consideration a dynamic behavior of the technical device (Martin, Para. 0023, 0032-0034, 0072 and 0094 – where the neural network used for creating a model can be a “recurrent neural network”, where the neural network is trained and adapted simultaneously, such that it is dynamic) wherein the neural network includes a fully connected output layer (Martin, Fig. 1 and Para. 0109 – where Fig. 1 shows the “neural network” having an “output layer”, where as shown on Fig. 1, is fully connected to the nodes of the other layers of the neural network).
PNG
media_image1.png
756
777
media_image1.png
Greyscale
Martin, Fig. 1
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the device including the above limitations of Schat in view of Cheng and Ohlarik to include wherein the data-based manipulation model includes a neural network that contains recurrent components that permit the data-based manipulation detection model to take into consideration a dynamic behavior of the technical device, as taught by Martin, in order to reduce the computational effort and runtime of the model (Martin, Para. 0037).
Schat in view of Cheng, Ohlarik, and Martin does not explicitly teach wherein the neural network includes an output layer that outputs the output vector.
However, Mahdi teaches wherein the neural network includes an output layer that outputs the output vector (Mahdi, Section I – where it is known in the art that recurrent neural networks (RNN) have the ability to “build on earlier types of networks with fixed-size input and output vectors”, where Mahdi proposes a “many-to-many” network structure with multiple layers which provides “output vectors” that predict “DOC upstream and downstream exhaust gas temperatures”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the device including the above limitations of Schat in view of Cheng, Ohlarik, and Martin, to include wherein the neural network includes an output layer that outputs the output vector, as taught by Mahdi, in order to provide a model which produces many outputs which reflect the state of the technical device.
In regards to Claim 10, Schat in view of Cheng, Ohlarik, Martin, and Mahdi teaches the device of Claim 9, and Schat further teaches wherein the technical device is an exhaust gas aftertreatment device in a motor vehicle (Schat, Para. 0001 and 0015 – where in one example, a defeat device is used to deactivate “a urea-based exhaust after-treatment system for a diesel vehicle”, such that it manipulates the after-treatment system, outside testing environments, to “provide better performance”).
Regarding Claim 11, Schat teaches: A non-transitory machine-readable memory medium on which is stored a computer program for manipulation detection of a technical device, the computer program, when executed by a computer (Schat, Abstract, Para. 0001 and 0020 – a method for “detecting defeat devices in an engine control unit”; where defeat devices are used to manipulate results of tests; where the method is executed by an ECU including a CPU in communication with a memory), causing the computer to perform the following steps:
providing an input vector including one or multiple system variables (Schat, Para. 0018, 0032-0033, and 0041 – “input data” that is input for a “machine-learning algorithm”, where the input data is indicative of “different input conditions” for testing; where the testing input conditions can be related to “engine temperatures”, speed changes, “air temperature”, “reduced cooling capacity of the cooling system”, fuel consumption, etc.; where the input data constitutes an input vector within a coded algorithm) and including at least one control variable for an intervention in the technical device, for successive time steps (Schat, Para. 0032-0033 and 0036-0042 – where input data is used to create a “test environment”, where input conditions include constants, or controls, which are set in the test, such as “a time constant within a certain time range”, a constant speed for several minutes, an air temperature staying constant, etc.; where the test is conducted over a period of time due to the behavior of defeat devices being “time dependent”);
using a data-based manipulation detection model (Schat, Para. 0025-0026 – “a machine-learning method” which accumulates knowledge of environmental test results to anticipate behavior of defeat devices) to generate a corresponding output vector where the machine learning includes an estimation methods, such that the machine learning method will determine, or output, which “data positions” and “time slots” have the highest likelihood “of changing the defeat device's behavior”, such that the likelihood is the output; where the output constitutes an output vector within a coded algorithm and where the machine learning is based on “input data”),
providing an actual monitored variable based on an at least one measured value in the successive time steps (Schat, Para. 0025-0026, 0032, and 0042 – where the machine learning method determines “data positions”, or an actual monitored variable, where the machine learning method is based on collected key-data, or measured values; where the data behavior is time dependent and the output includes “time slots”);
creating a where the machine learning method estimates a highest likelihood of a defeat device changing behavior; where the likelihood is the classification and the data used is time dependent); and
detecting a manipulation as a function of the measurement classification vector and a first and a second comparison vector for one or multiple of the time steps of a time window (Schat, Para. 0042 – where a comparison of key-data and ECU data “will indicate the presence of a defeat device, if there is a difference in the comparison (e.g., an anomaly)”), the first and the second comparison vector being determined by rounding element values of the output vector based on a first manipulation threshold value and a second manipulation threshold value, which is different from the first manipulation threshold value, as rounding limits (Schat, Para. 0042 – where “key-data is considered to be “similar” when changes are due merely to uncontrollable random variations or the input data and internal states. In one embodiment, a level of these random variations is predetermined by adding a threshold value below which a pair of key-data is considered to be similar.”), wherein a controller configured to control, based on the control variable, a physical action of the technical device (Schat, Para. 0001, 0015, 0032-0033 and 0036-0042 – where a “urea-based exhaust after-treatment system” is activated or deactivated based on when a “defeat device” detects an “environmental test condition”; where the environmental test condition set by input data used to create a “test environment”, where input conditions include constants, or controls, which are set in the test).
While Schat teaches using a data-based manipulation detection model to generate a corresponding output vector, the data-based manipulation detection model being configured to output an output vector, for the input vector, Schat does not fully teach outputting an output vector as a classification vector, which indicates a classification of a monitored variable in value ranges, for the input vector. Additionally, while Schat teaches creating an output vector from the actual monitored variable for each time step, Schat does not teach creating a measurement classification vector from the actual monitored variable for each time step, and wherein the data-based manipulation model includes a neural network that contains recurrent components that permit the data-based manipulation detection model to take into consideration a dynamic behavior of the technical device, wherein the neural network includes a fully connected output layer that outputs the output vector.
However, Cheng teaches outputting an output vector as a classification vector, which indicates a classification of a monitored variable in value ranges, for the input vector (Cheng, 2. Method – where a machine learning model receives an input vector and outputs a “probability distribution vector” and considers the vector as “a cumulative probability distribution on categories (1, ..., k, ..., K)”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the non-transitory machine readable medium of Schat to include outputting an output vector as a classification vector, which indicates a classification of a monitored variable in value ranges, for the input vector, as taught by Cheng, in order to allow the machine learning model to train on larger data sets and make faster predictions once trained on the categorized data (Cheng, Abstract and 4. Discussion and Future Work).
Schat in view of Cheng does not teach creating a measurement classification vector from the actual monitored variable for each time step, and wherein the data-based manipulation model includes a neural network that contains recurrent components that permit the data-based manipulation detection model to take into consideration a dynamic behavior of the technical device, wherein the neural network includes a fully connected output layer that outputs the output vector.
However, Ohlarik teaches creating a measurement classification vector from the actual monitored variable for each time step (Ohlarik, Para. 0067-0069 – where a machine learning model takes in vectors which represent changes in a measurement, in this case location, for “consecutive time intervals”, and generates an output vector which represents a likelihood of a collision “occurring during a particular time interval”, where the classes of the vector represent specific time steps such as 0.1 seconds, 0.2 seconds, etc.; where the likelihood is the classification).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the non-transitory machine readable medium of Schat in view of Cheng to include creating a measurement classification vector from the actual monitored variable for each time step, as taught by Ohlarik, in order to measure the changes within the technical device over time steps to determine if there is manipulation occurring over time.
Schat in view of Cheng and Ohlarik does not teach wherein the data-based manipulation model includes a neural network that contains recurrent components that permit the data-based manipulation detection model to take into consideration a dynamic behavior of the technical device wherein the neural network includes a fully connected output layer that outputs the output vector.
However, Ohlarik teaches creating a measurement classification vector from the actual monitored variable for each time step (Ohlarik, Para. 0067-0069 – where a machine learning model takes in vectors which represent changes in a measurement, in this case location, for “consecutive time intervals”, and generates an output vector which represents a likelihood of a collision “occurring during a particular time interval”, where the classes of the vector represent specific time steps such as 0.1 seconds, 0.2 seconds, etc.; where the likelihood is the classification).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the device of Schat in view of Cheng to include creating a measurement classification vector from the actual monitored variable for each time step, as taught by Ohlarik, in order to measure the changes within the technical device over time steps to determine if there is manipulation occurring over time.
Schat in view of Cheng and Ohlarik does not teach wherein the data-based manipulation model includes a neural network that contains recurrent components that permit the data-based manipulation detection model to take into consideration a dynamic behavior of the technical device, wherein the neural network includes a fully connected output layer that outputs the output vector.
However, Martin teaches wherein the data-based manipulation model includes a neural network (Martin, Para. 0003-0004 and 0023 – using “artificial neural networks” to create a “mathematical model” representative of a “exhaust gas aftertreatment component of an exhaust gas aftertreatment system”) that contains recurrent components that permit the data-based manipulation detection model to take into consideration a dynamic behavior of the technical device (Martin, Para. 0023, 0032-0034, 0072 and 0094 – where the neural network used for creating a model can be a “recurrent neural network”, where the neural network is trained and adapted simultaneously, such that it is dynamic) wherein the neural network includes a fully connected output layer (Martin, Fig. 1 and Para. 0109 – where Fig. 1 shows the “neural network” having an “output layer”, where as shown on Fig. 1, is fully connected to the nodes of the other layers of the neural network).
PNG
media_image1.png
756
777
media_image1.png
Greyscale
Martin, Fig. 1
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the non-transitory computer readable medium including the above limitations of Schat in view of Cheng and Ohlarik to include wherein the data-based manipulation model includes a neural network that contains recurrent components that permit the data-based manipulation detection model to take into consideration a dynamic behavior of the technical device, as taught by Martin, in order to reduce the computational effort and runtime of the model (Martin, Para. 0037).
Schat in view of Cheng, Ohlarik, and Martin does not explicitly teach wherein the neural network includes an output layer that outputs the output vector.
However, Mahdi teaches wherein the neural network includes an output layer that outputs the output vector (Mahdi, Section I – where it is known in the art that recurrent neural networks (RNN) have the ability to “build on earlier types of networks with fixed-size input and output vectors”, where Mahdi proposes a “many-to-many” network structure with multiple layers which provides “output vectors” that predict “DOC upstream and downstream exhaust gas temperatures”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the non-transitory computer readable medium including the above limitations of Schat in view of Cheng, Ohlarik, and Martin, to include wherein the neural network includes an output layer that outputs the output vector, as taught by Mahdi, in order to provide a model which produces many outputs which reflect the state of the technical device.
In regards to Claim 12, Schat in view of Cheng, Ohlarik, Martin, and Mahdi teaches the method of Claim 1, and Schat in view of Cheng, Ohlarik, Martin, and Mahdi further teaches wherein the fully connected output layer (Martin, Fig. 1 and Para. 0109 – where Fig. 1 shows the “neural network” having an “output layer”, where as shown on Fig. 1, is fully connected to the nodes of the other layers of the neural network) includes a monotonously increasing activation function (Cheng, Section 2 – “output probability vector (o1, ...ok, ...oK)”, such that the output vector outputted by the “output layer” is monotonously increasing i.e. “(o1, o2, ..., ok, ...oK)”, where the output transfer function used is a “sigmoid function” which “sets the target value of output nodes Oi (i ≤ k) to 1 and Oi (i > k) to 0”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the method including the above limitations of Schat in view of Cheng, Ohlarik, Martin, and Mahdi to further include wherein the fully connected output layer includes a monotonously increasing activation function, as taught by Cheng, in order to improve training consistency and accuracy.
In regards to Claim 13, Schat in view of Cheng, Ohlarik, Martin, and Mahdi teaches the device of Claim 9, and Schat in view of Cheng, Ohlarik, Martin, and Mahdi further teaches wherein the fully connected output layer (Martin, Fig. 1 and Para. 0109 – where Fig. 1 shows the “neural network” having an “output layer”, where as shown on Fig. 1, is fully connected to the nodes of the other layers of the neural network) includes a monotonously increasing activation function (Cheng, Section 2 – “output probability vector (o1, ...ok, ...oK)”,