Real-time machine learning-based IMU correction

Abstract

The presently disclosed subject matter includes a system and method dedicated to mitigating inertial measurement unit (IMU) axis failures. The system comprises processing circuitry operatively connectable to an IMU including accelerometer and rate gyro subsystems configured to provide real-time IMU data along multiple axes. The processing circuitry is configured to monitor the real-time IMU data and detect a fault affecting at least one IMU axis. Responsive to detecting the fault, a machine learning model trained on relationships between IMU measurements is applied to reconstruct missing or degraded accelerometer or rate gyro data based on remaining available IMU data, thereby generating reconstructed IMU data representative of platform motion.

Description

[0001] REAL-TIME MACHINE LEARNING-BASED IMU CORRECTION TECHNICAL FIELD

[0002] The presently disclosed subject matter relates to augmentation of inertial navigation.

[0003] BACKGROUND

[0004] Inertial Measurement Units (IMUs) are critical components in aerospace navigation and control systems, while also finding applications in various other motionsensing fields. These devices typically comprise multiple sensors designed to measure motion across six degrees of freedom.

[0005] In standard IMU architectures, two main core subsystems are employed: a triaxial accelerometer and a triaxial rate gyro. The accelerometer subsystem includes three sensors, each aligned along orthogonal axes (x, y, z) to measure linear acceleration. These measurements are crucial for detecting the translation of aircraft, spacecraft, or other aerospace platforms, and are typically expressed in gravitational acceleration (g) or meters per second squared (m / s2). Similarly, the rate gyro subsystem contains three sensors, each oriented to capture angular velocity around the same orthogonal axes (x, y, z). The rate gyro outputs angular velocity measurements are typically expressed in degrees per second (° / s) or radians per second (rad / s). By combining these six axes of measurement, the IMU provides comprehensive motion tracking essential for platform navigation and control.

[0006] GENERAL DESCRIPTION

[0007] Despite their utility, existing IMU devices suffer from various technical challenges. One significant problem arises when sensors along one or more axes become nonfunctional or provide degraded measurements. These failures can result, for example, from mechanical stress, environmental conditions, or electronic component failure,leading to compromised motion tracking accuracy and system reliability. This problem is prominent in MEMS-based IMUs, which, despite advantages in size, cost, and power consumption, are especially vulnerable to environmental stresses.

[0008] The presently disclosed subject matter provides a novel solution for IMU axis failures. The proposed solution leverages a machine learning model trained to compensate for IMU sensor failures. The model is trained to process real-time inputs from functioning axes and estimate missing or degraded IMU measurements, enabling motion tracking, even when certain axes are compromised. This approach enables real-time fault correction, enhancing the robustness of IMU-based systems in demanding environments.

[0009] According to a first aspect of the presently disclosed subject matter there is provided an IMU (Inertial Measurement Unit) error mitigation circuitry operatively connectable to an IMU mounted / mountable onboard a platform; wherein the IMU comprises an accelerometer subsystem configured to measure linear acceleration along three orthogonal axes (x, y, z) and a rate gyro subsystem configured to measure angular velocity around three orthogonal axes (x, y, z); the accelerometer subsystem and the rate gyro subsystem being configured to provide real-time IMU data;

[0010] the IMU error-mitigation circuitry is configured to:

[0011] monitor the real-time IMU data to detect an IMU fault in the real-time IMU data; responsive to detecting an IMU fault in at least one IMU axis, apply a machine learning model trained to process the real-time IMU data, including accelerometer and rate gyro measurements, and reconstruct missing or degraded data resulting from the IMU fault to obtain reconstructed IMU data;

[0012] wherein the machine learning model is trained to reconstruct missing or faulty IMU data for the at least one IMU axis based on partially available IMU data by leveraging learned relationships between IMU measurements, thereby generating complete IMU data that reflects the dynamic movement of the platform; and

[0013] provide reconstructed IMU data to a client.In addition to the above features, the method according to this aspect of the presently disclosed subject matter can optionally comprise one or more of features (i) to (x) below, in any technically possible and technically possible combination or permutation:

[0014] i. Wherein the machine learning model is trained using a training dataset comprising IMU data series, each IMU data series being a time-series of IMU measurements gathered over a respective trajectory.

[0015] ii. Wherein the IMU data series in the training dataset corresponds to platforms of a type matching that of the platform on which the IMU is mounted, ensuring that the machine learning model can address distinct motion characteristics inherent to the platform.

[0016] iii. Wherein the IMU error-mitigation circuitry is configured to provide the reconstructed IMU data to a navigation computer onboard the platform to facilitate navigation of the platform.

[0017] iv. Wherein the platform is any one of: ground vehicle, aerospace vehicle, and a marine vehicle.

[0018] v. Wherein the platform is an aerospace vehicle, and the processing circuitry is configured to generate the reconstructed IMU data during flight to enable continuous motion tracking of the platform in the presence of an IMU failure.

[0019] vi. Wherein the IMU failure is a failure of at least one of: a rate gyro sensor in the rate gyro subsystem configured to measure angular velocity around a respective axis; and an accelerometer sensor in the accelerometer subsystem configured to measure linear acceleration along a respective axis.

[0020] vii. Wherein the IMU data processed by the machine learning model includes, in addition to the accelerometer and rate gyro measurements, auxiliary data generated by another sensor or device onboard the platform.viii. The IMU error-mitigation circuitry further comprises a magnetometer configured to measure the magnetic field in three axes, wherein the auxiliary data includes magnetic field measurements obtained by the magnetometer.

[0021] ix. Wherein activation of the machine learning model includes applying the machine learning model on IMU output obtained from only viable IMU sensors.

[0022] x. Where the processing circuitry is configured to apply a lever arm correction to the accelerometer and rate gyro measurements before applying the machine learning model, thereby enabling the application of the same machine learning model for reconstructing IMU data across different mounting placements of the IMU on the platform.

[0023] According to a second aspect there is provided a computer-implemented method of real-time fault-mitigation in an Inertial Measurement Unit (IMU) mounted on a platform, the method can be executed by a processing circuitry operatively connected to the IMU; the method comprising:

[0024] obtaining real-time IMU data, comprising:

[0025] obtaining linear acceleration measured along three orthogonal axes (x, y, z) using an accelerometer subsystem;

[0026] obtaining angular velocity measured around three orthogonal axes (x, y, z) using a rate gyro subsystem;

[0027] monitoring the real-time IMU data;

[0028] detecting an IMU fault in at least one IMU axis of the real-time IMU data; activating a machine learning model trained to process the real-time IMU data, including accelerometer and rate gyro measurements, and reconstructing missing or degraded data resulting from the IMU fault; wherein the machine learning model is trained to reconstruct missing or faulty IMU data for at least one IMU axis based on partially available IMU data by leveraging learned relationships between different IMUmeasurements and thereby generating complete IMU data that reflects dynamics of the platform; and providing reconstructed IMU data to a client.

[0029] According to a third aspect there is provided an Inertial Navigation System mountable on a platform and comprising or otherwise operatively connected to the IMU disclosed according to the first aspect.

[0030] According to a fourth aspect there is provided a platform comprising the IMU according to the first aspect.

[0031] According to a fifth aspect there is provided a non-transitory computer-readable medium comprising instructions that, when executed by a computer, cause the computer to perform a method of real-time fault-mitigation in an Inertial Measurement Unit (IMU) mounted on a platform, the method comprising:

[0032] obtaining real-time IMU data, the real-time IMU data comprising:

[0033] linear acceleration measurements along three orthogonal axes (x, y, z) received from an accelerometer subsystem; and

[0034] angular velocity measurements around three orthogonal axes (x, y, z) received from a rate gyro subsystem;

[0035] monitoring the real-time IMU data to detect an IMU fault in at least one IMU axis; detecting an IMU fault in at least one IMU axis of the real-time IMU data; responsive to detecting the IMU fault, activating a machine learning model trained to process the real-time IMU data, including accelerometer and rate gyro measurements, and reconstructing missing or degraded data resulting from the IMU fault;

[0036] wherein the machine learning model is trained to reconstruct missing or faulty IMU data for at least one IMU axis based on partially available IMU data by leveraging learned relationships between different IMU measurements and thereby generating complete IMU data that reflects the dynamics of the platform; and providing reconstructed IMU data to a client.The non-transitory computer-readable medium can be installed in processing circuitry of an IMU device or a processing circuitry connectable to an IMU device to enable to carry out a continuous motion-tracking even in the presence of failed IMU-data.

[0037] According to a sixth aspect of the presently disclosed subject matter there is provided an Inertial Measurement Unit (IMU) with an internally integrated circuitry providing real-time IMU fault mitigation capabilities; the IMU comprising:

[0038] an accelerometer subsystem configured to measure linear acceleration along three orthogonal axes (x, y, z) and a rate gyro subsystem configured to measure angular velocity around three orthogonal axes (x, y, z); the accelerometer subsystem and the rate gyro subsystem being configured to provide real-time IMU data;

[0039] the circuitry is configured to:

[0040] monitor the real-time IMU data to detect an IMU fault in the real-time IMU data; responsive to detecting an IMU fault in at least one IMU axis, apply a machine learning model trained to process the real-time IMU data, including accelerometer and rate gyro measurements, and reconstruct missing or degraded data resulting from the IMU fault to obtain reconstructed IMU data;

[0041] wherein the machine learning model is trained to reconstruct missing or faulty IMU data for the at least one IMU axis based on partially available IMU data by leveraging learned relationships between IMU measurements, thereby generating complete IMU data that reflects the dynamic movement of the platform; and

[0042] provide reconstructed IMU data to a client.

[0043] The method, the INS, the platform, the non-transitory program storage device, and the IMU with internally integrated circuitry disclosed above, can optionally comprise one or more of features (i) to (x) listed above, mutatis mutandis, in any technically possible combination or permutation.BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to understand the presently disclosed subject matter and to see how it may be carried out in practice, the subject matter will now be described, by way of nonlimiting examples only, with reference to the accompanying drawings, in which:

[0045] FIG. 1 is a schematic block diagram of an IMU error-mitigation circuitry, according to some examples of the presently disclosed subject matter;

[0046] FIG. 2 is a schematic block diagram of an INS comprising the IMU error-mitigation circuitry shown in FIG. 1, according to some examples of the presently disclosed subject matter;

[0047] Fig.3 is a flowchart illustrating operations carried out during operation of the IMU for real-time fault mitigation, according to some examples of the presently disclosed subject matter;

[0048] Fig. 4 is a schematic block diagram of a computer system configured to train the IMU Fault mitigation ML model, according to some examples of the presently disclosed subject matter; and

[0049] Fig. 5 is a flowchart illustrating operations carried out for training the IMU Fault mitigation ML model, according to some examples of the presently disclosed subject matter.

[0050] DETAILED DESCRIPTION

[0051] As used herein, the term "platform" broadly refers to any body, device, structure, or vehicle that carries the disclosed IMU error-mitigation circuitry 150. This includes, but is not limited to, airborne vehicles (e.g., missiles, aircraft, drones, satellites), ground vehicles, and marine vehicles (e.g., ships, boats, submarines).

[0052] When a platform moves through an environment— whether in space, on land, in water, or in the air— it follows a trajectory determined by its physical properties (e.g., mass distribution, geometric configuration, control-surface configurations, and propulsioncharacteristics) and environmental conditions (e.g., gravitational field, magnetic field, medium density, and initial state parameters). These properties and conditions define a set of possible trajectories that share common characteristics. Each trajectory within this set represents a solution to the equations of motion under specific initial conditions and parameter values, while maintaining the characteristic dynamic patterns inherent to the platform.

[0053] Each trajectory has an associated IMU data series, which is the specific time-series of IMU measurements gathered along the trajectory through acceleration and angular velocity readings (a matrix of the measurement vectors over time). While these IMU data series are unique to each specific trajectory, they exhibit characteristic inertial patterns that reflect the platform's inherent dynamic behavior across its set of possible trajectories. For example, in the case of an aerospace body (e.g., missiles, satellites, shells, mortars, rockets, and both manned and unmanned aircraft), the IMU readings provide insights into the accelerations and rotational dynamics experienced along its path. These readings highlight both the distinctive features of its movement and the common dynamic patterns shared by all possible trajectories for that type of platform.

[0054] The presently disclosed subject matter includes a novel solution for compensating for the failure of one or more IMU axes. This solution leverages a machine learning model trained using a training dataset comprising a collection of IMU data series of a certain type of platform (e.g., a certain type of missile or satellite). The model enables reconstruction of missing or degraded data resulting from a failed IMU sensor of a respective axis (or more than one), effectively replacing the missing component, thereby ensuring continuous, reliable motion tracking. This approach strengthens IMU-based systems by improving their robustness and precision, making them more suitable for critical applications where performance must remain consistent. The innovation benefits MEMS IMUs, as these compact sensors are more vulnerable to axis failures when exposed to harsh environments.

[0055] Attention is now drawn to FIGS. 1 and 2, which present schematic block diagramsof a system 100 designed to mitigate IMU faults in real time, according to examples of the present disclosure. FIG. 1 illustrates system 100 comprising an IMU 110 and an IMU errormitigation circuitry 150, while FIG. 2 depicts an Inertial Navigation System (INS) that integrates the IMU and the IMU error-mitigation circuitry. These figures are provided solely for illustrative purposes, and the specific configurations shown should not be interpreted as limiting in any way.

[0056] Both system 100 shown in in FIG. 1 and the INS 120 shown in FIG. 2 include at least one processing circuitry configured to execute tasks specific to their respective functions, including IMU fault detection and IMU fault mitigation. The processing circuitry may include one or more processors coupled to computer-memory devices to execute the functions of each module. In some configurations, the processing circuitry is configured to execute multiple functional modules based on computer-readable instructions stored in a non-transitory memory device. Such functional modules are herein described as being comprised within the processing circuitry. For example, IMU error-mitigation circuitry 150 executes the Fault Detection Module (115) and the Fault-Mitigating ML model (117) described below.

[0057] In FIG. 1, system 100 shows an IMU 110 operatively connected to IMU errormitigation circuitry 150 configured to receive IMU measurements and reconstruct IMU data in case of an IMU fault. IMU (110) includes the two primary sensor subsystems: an accelerometer (111) and a rate gyro (113). As mentioned above, the accelerometer is configured to measure linear acceleration along three orthogonal axes (x, y, z), while the rate gyro is configured to measure angular velocity around these three axes (x, y, z).

[0058] To enhance fault tolerance, the current disclosure introduces additional components. Specifically, circuitry 150, operatively connected to the IMU 110 comprises a Fault Detection Module (115), which is configured for real-time monitoring of faults involving one or more of the six IMUs sensor components. Upon detecting a fault, the system activates a Fault-Mitigating ML model (117), designed to reconstruct the missing or degraded data resulting from the failure in IMU operation. The output of the machinelearning model is reconstructed IMU data (119), which can be stored in computer memory and / or provided directly for further processing e.g., in subsequent navigation or analytical modules. In some examples, if the Fault Detection Module (115) detects no faults, the raw IMU data is used as is for further processing.

[0059] It is noted that the disclosed subject matter contemplates two possible implementations for IMU error mitigation.

[0060] In the first implementation, an IMU error-mitigation circuitry is connected to an existing IMU via an external connection. This configuration allows the external circuitry to receive data from the IMU, detect errors, and apply mitigation measures in real time. By enablingthe use of error-mitigation circuitry with standard IMUs, this approach eliminates the need for modifications to the IMU's internal design. This flexibility allows existing IMUs to be retrofitted with advanced error-mitigation capabilities.

[0061] In the second implementation, an IMU is designed with integrated error-mitigation functionality. This approach incorporates internal IMU circuitry specifically configured to detect and mitigate errors directly within the IMU, creating a self-contained solution. By embedding error-mitigation capabilities within the IMU itself.

[0062] In FIG. 2, system 100 depicted in FIG. 1 is integrated within an inertial navigation system (INS). In this configuration, the IMU's reconstructed IMU data (119) serves as input to additional INS modules.

[0063] An INS commonly requires initial conditions and a navigation datum as basic inputs. Initial conditions include the starting position, velocity, and orientation of the system, which serve as the baseline for all subsequent calculations. The navigation datum, such as the Earth-Centered WGS84, provides a global reference framework, ensuring that all positional and navigational outputs are consistent with the Earth's geometry and rotation.

[0064] The Navigation Equations Module (125) processes updated IMU data, along with compensation inputs, to calculate attitude, velocity, and position updates for the vehicle.It should be noted that, while IMU error-mitigation circuitry 150 depicted in FIG. 2 is used within the context of an INS to support the broader navigation solution, the IMU system is configured and can be used to provide data to various client devices beyond an INS, demonstrating its adaptability and modularity. Likewise, IMU error-mitigation circuitry can be connected to IMU used in various devices. It is well-suited for diverse applications in fields such as consumer electronics, UAV stabilization, and wearable technology. The inclusion of the system within an INS in FIG. 2 is solely provided as an example and should not be construed as limiting.

[0065] The IMU error-mitigation circuitry 150 shown in figure 1 or INS shown in figure 2, can be installed within a platform, including various types of ground, marine and aerospace vehicles, to provide reliable IMU output even in the event of a sensor failure. Examples of such platforms include projectiles with simple trajectories, such as mortars, and those with complex, maneuvering trajectories, such as guided missiles. Additionally, the IMU error-mitigation circuitry 150 and INS are suitable for integration into other aerospace applications, such as rockets, satellites, and unmanned aerial vehicles (UAVs), where precise motion sensing is essential for navigation, stability, and mission success.

[0066] For illustrative purposes, consider an example where the IMU error-mitigation circuitry 150 is installed on a satellite, by connecting it to a satellite IMU. In this context, the IMU's role is critical for maintaining precise orientation and stability in orbit. As the satellite experiences various forces, such as gravitational pull and potential vibrations from internal mechanisms, the IMU continuously senses acceleration and angular velocity to help manage the satellite's attitude and control systems. Once the IMU error-mitigation circuitry 150 completes its data processing, the reconstructed IMU data (119), if applicable, is provided to other navigation or control modules within the satellite's onboard systems, enabling real-time adjustments to its position, velocity, and orientation. The fault detection and IMU-data reconstruction capabilities are advantageous in this scenario, as they enable to deliver reliable data even in the event of IMU sensor failure or degradation, thereby supporting the satellite's long-term operability in orbit.FIG. 3 is a flowchart of operations carried out by IMU error-mitigation circuitry 150, according to some examples of the presently disclosed subject matter. As mentioned, IMU error-mitigation circuitry 150 can be installed in a platform to provide continuous, realtime acceleration and rotational rates measurements, while also performing real-time (e.g., in-flight) detection and reconstruction of missing or degraded measurements.

[0067] At block 301, IMU output are obtained (e.g., from IMU 110), including linear acceleration along three orthogonal axes (x, y, z) and angular velocity around these three axes (x, y, z). This data, provided by the accelerometer and rate gyro components within the IMU, represents the core IMU measurements of motion and orientation.

[0068] As mentioned above, IMU error-mitigation circuitry 150 is configured to implement a machine learning model that detects and corrects missing or degraded IMU measurements in real-time. Operations carried out by the processing circuitry are described below.

[0069] At block 305, the IMU outputs are actively monitored in real-time (e.g., by Fault Detection module 115) to detect any potential faults in the accelerometer, rate gyro, or related circuitry. Methods for fault identification in an IMU may vary, depending on requirements, hardware, or the stage of system operation. Examples of such methods include:

[0070] Built-In Tests (BITs): Some components include self-test mechanisms that monitor their own health status and send one or more health status bits to the upper-level system. For example, Fault Detection module 115 can be configured to receive a status bit and act accordingly.

[0071] Basic Output Tests for IMUs: The data from IMU can be examined, and simple tests such as verifying the regular progression of the clock or real-time data updates can be performed. For example, Fault Detection module 115 can be configured to ensure that there are no sudden stops or unexpected delays in the data stream.

[0072] Axis Variance Test: Fault Detection module 115 can be configured to calculate thevariance of readings along a specific axis. If the variance remains close to zero over time, it may indicate that the axis is inactive or malfunctioning.

[0073] Boundary and Anomalous Value Tests: Fault Detection module 115 can be configured to examine the values received from the measurement unit: if a specific axis exceeds reasonable limits, reaches saturation without an apparent cause, or produces physically impossible values, this may indicate a fault.

[0074] If no faults are detected, the IMU output measurements are transmitted at block 309 for subsequent processing by the relevant client systems (e.g., INS), as applicable.

[0075] If a fault is detected, the process moves to block 307, where the Fault-Mitigating ML model is activated. During inference, the model uses IMU data series— obtained in real- time from the IMU and containing measurements such as accelerations and rate gyro outputs along the available axes— to reconstruct missing or degraded sensor outputs. In some examples, if a sensor (of a specific accelerometer or gyroscope axis) is identified as faulty, only the outputs of the viable sensors are provided to the ML model, while data from the faulty sensor is excluded. In other examples, IMU data series from all six sensors are provided to the ML model. As further explained in FIG. 5, since the model has already learned, during training, the relationships between sensor readings by analyzing characteristic inertial patterns of the IMU data series, it can estimate the missing sensor data using the available IMU data series.

[0076] The reconstructed data is then passed forward, at block 309, for subsequent processing by client systems (e.g., INS) thus ensuring continuity of IMU operability, despite the detected fault.

[0077] In some examples, the machine learning model can be trained to use, in addition to the IMU data, information from other sensors such as a magnetometer. This auxiliary data, when used during inference, can help increase the robustness and accuracy of the model. The use of such information depends on its availability during inference, for instance the presence of a magnetometer onboard the platform.If auxiliary data is used in response to the detection of an IMU failure, it is obtained from the relevant sensor (e.g., magnetic field measurements obtained from a three-axis IMU magnetometer) and provided as input to the Fault-Mitigation Model 117 in addition to the acceleration and rate gyro IMU data and applied to the Fault-Mitigating ML model.

[0078] FIG. 4 is a block diagram of a computer system (400) dedicated to training a Fault-Mitigation ML model (117), according to some examples of the presently disclosed subject matter. The computer system can comprise for example at least one processing circuitry configured to perform operations dedicated for ML training. A processing circuitry may include one or more processors in conjunction with computer-memory devices to execute different modules. For instance, a processing circuitry 410 can execute functional modules including Trajectory Generator (411), IMU Simulator (413), and Machine Learning Training module (415).

[0079] As explained above, each platform is characterized by a respective IMU data series that exhibits characteristic patterns reflecting its inherent dynamic behavior within its range of possible trajectories. Accordingly, the model is specifically trained for a particular type of platform (e.g., a specific type of missile, satellite, or ground vehicle), allowing for the generation of distinct models, each tailored to different types of patterns. Each ML model, dedicated to being used in a certain type of platform, is trained using specific training data derived from the IMU data series of that platform type. This ensures that the model's Fault-Mitigation capabilities can address the distinct dynamics and motion characteristics inherent to the platform.

[0080] FIG. 5 is a flowchart of operations carried out for training the Fault-Mitigation ML model, according to some examples of the presently disclosed subject matter. By way of example only, operations in FIG. 5 are described with reference to components in FIG. 4. Training of the model can be executed on a computer system, and the trained model can be installed in an IMU after training.

[0081] At block 501, a collection of nominal simulated trajectories for a certain type of platform is obtained (e.g., by Trajectory Generator 411). These trajectories can beobtained by simulating platform motion according to equations of motion under the model's operating conditions. For the purposes of this application, platforms are considered to be of the ’same type’ if they possess sufficiently similar dynamic, physical, inertial, and geometric properties that yield comparable dynamic behavior under the model’s operating conditions, regardless of their formal classifications or conventional categorizations. This similarity includes attributes such as mass distribution, response to applied forces, and performance parameters within the specified operational envelope. For example, in this context, two projectiles of the same nominal model may be considered as belonging to a different type if variations in payload placement (e.g., the location of the IMU system) alter their dynamic response. Notably, lever arm correction can be applied to compensate for different IMU placements across platforms. In some examples, when lever arm correction is applicable, platforms with different sensor placements are considered to be of the same type. A lever arm is the vector offset between the IMU’s sensing elements and a defined reference point on the platform, accounting for both distance and direction in the platform's coordinate frame. In some examples, IMU error-mitigation circuitry 150 includes a lever arm correction module (not shown) that applies corrections to the measured acceleration and angular rate data, thereby enabling the application of the same model for different IMU placements.

[0082] To more accurately simulate the trajectory of a platform, several parameters are defined, including, for example, the initial position, velocity, and the body's inertial properties, as well as external environmental forces such as gravity and aerodynamic drag. Using these parameters, mathematical techniques such as numerical integration, can iteratively compute the platform's position and velocity throughout its trajectory. For example, by adjusting variables such as launch angle, initial speed, and environmental factors (e.g., wind conditions), a comprehensive set of trajectories for the same type of platform can be generated. This set captures a wide range of possible trajectories (e.g., aerial vehicle flight paths) under varying conditions. In some implementations, synthetic trajectories may be supplemented or replaced by historical data from recorded trajectories during training.Respective IMU data series are generated for the simulated trajectories. Given the collection of trajectories, a respective IMU data series can be generated for each trajectory, incorporating the specific physical properties of the platform, such as mass, center of mass, and moments of inertia. These properties, combined with the dynamics of each unique trajectory, determine the expected IMU measurements, including linear acceleration and angular velocity, that would be recorded along the trajectory. The platform's mass affects linear acceleration under applied forces, while its center of mass and moments of inertia influence rotational behavior and angular velocity in response to torques. This IMU data series thus captures how the platform's specific physical characteristics modulate its response to forces encountered at each point along the trajectory, enabling the simulation of IMU data as it would be recorded under real-world conditions for each distinct trajectory.

[0083] At block 503, in some examples the nominal trajectories and their respective IMU data series are modified to more accurately simulate real-world data, ensuring a realistic basis for training the model. One such modification involves noise injection, which is applied to the IMU data series to reflect the inherent imperfections and variability in real-world IMU measurements (e.g., by an IMU simulator 413, configured to simulate, for example, real flight measurements). This noise can include random sensor noise, bias, drift, and environmental disturbances, each of which affects the accuracy of IMU readings in practice. By incorporating these noise components into the profiles, the simulated IMU data better captures the variations and imperfections observed in real-world measurements, providing a more robust and realistic dataset for model training.

[0084] At block 505, the IMU data series are processed to remove one (or optionally more) of the six distinct time-series vectors corresponding to the six IMU sensors (three accelerometers and three rate gyros) for each trajectory. This removal simulates missing or degraded IMU data, which can be used for training purposes (e.g., by the Training Data Preparation Module 415).

[0085] In one example, data from one of the rate gyro sensors is removed. Removing thedata from a rate gyro sensor— for example, the x-axis rate sensor— introduces a realistic scenario where rotational data for that axis is unavailable. This missing data scenario allows the system to test and train the model's ability to reconstruct or infer the removed angular velocity data. In some examples, where IMU output of all six sensors is provided to the model during inference, the faulty sensor is not simulated by removal but by incorporating degraded data into the corresponding sensor.

[0086] In some implementations, the removed data serves as the ground truth for validation, allowing the model to compare its reconstructed output to the actual measurements. In other cases, the full set of six time-series vectors is used as the ground truth.

[0087] As mentioned above, in some examples the machine learning model is trained to use, in addition to the IMU data, information from other sensors such as a magnetometer. In such cases, during training, the relevant auxiliary information is incorporated into the training data to enhance the model’s performance. The IMU data series of simulated trajectories can also include magnetic field data if the simulation incorporates a modeled magnetic field, providing a more comprehensive representation of sensor inputs. This simulated magnetic field data can be used during training to improve the model’s accuracy and robustness, similar to how real-world magnetic field measurements would be used.

[0088] At block 507, the Fault-Mitigation machine learning model (117) is trained to enable it to reconstruct missing or degraded IMU measurements resulting from a faulty sensor or sensors (e.g., by Fault-Mitigation Training Module 417). The model can be a Deep Neural Network (DNN), such as a Long Short-Term Memory (LSTM), Transformer Model, or Temporal Convolutional Network (TCN).

[0089] During the training phase, the model is trained to predict missing IMU data by processing IMU data series and identifying relationships between IMU measurements generated by different sensors. The model learns to recognize characteristic inertial patterns inherent to the platform's movement across various trajectories, which represent recurring dynamic behaviors of a platform. Using a large dataset of IMU data seriesgathered from different scenarios— each containing measurements such as acceleration and angular velocity— the model establishes an understanding of how these sensors provide a representation of the overall movement. By leveraging these learned relationships, the model can reconstruct missing or faulty sensor data based on partial available IMU data, enabling the generation of a complete IMU data series that reflects the platform's dynamic movement.

[0090] Throughout the training process, the model's ability to predict missing or degraded data is continually evaluated, allowing iterative adjustments that refine its performance. For instance, a supervised learning approach could be used, where the model is trained using complete datasets as ground truth to evaluate and improve its predictions. During inference, the model leverages the learned characteristic inertial patterns to reconstruct missing data, ensuring reliable estimations even in the presence of incomplete datasets. This capability supports robust Fault-Mitigation during inference.

[0091] Once the training of the Fault-Mitigation model is finalized, it can be deployed for use with the IMU 110 (block 509). Since the model is specifically trained for a particular type of platform (e.g., a certain type of aerospace platform such as a missile or satellite), the trained model can be integrated with an IMU designated for use in that specific platform type, ensuring optimized fault tolerance and reliable data recovery in line with the platform's unique motion characteristics. For example, the trained model can be stored on a computer memory device integrated in IMU error-mitigation circuitry 150 which can be connected to or integrated with the IMU 110 (as shown in FIG. 1 and FIG. 2), which can then be installed in the relevant platform. Compatibility across different platforms may be possible depending on the similarity between their properties, such as mass distribution, control characteristics, and the required accuracy of the predictions. As mentioned above, a lever arm module can be used for lever arm correction applied to compensate for different IMU placements across platforms.

[0092] In some examples, to confirm the accuracy of the model’s output, the complete IMU measurements generated by the model are compared to the nominal IMU data seriesproduced by the IMU data series Generator. A difference smaller than a certain threshold indicates acceptable accuracy, confirming that the model's predictions align closely with the expected nominal profiles.

[0093] As discussed above, using the trained model in real-time during operation of the platform (e.g., during flight) as a backup to compensate for a failing sensor ensures continuous, reliable IMU measurements by dynamically predicting and replacing missing or degraded data, maintaining the integrity of navigation and control systems, even under fault conditions.

[0094] It is noted that the present application is not limited to the use of an IMU in navigation or INS applications. IMU technology finds diverse applications across multiple technical domains, including motion tracking in robotic systems, stabilization systems for aerial vehicles, gesture recognition interfaces, and vibration monitoring in structural health systems. Such applications, as well as others, are contemplated to be within the scope of the present disclosure.

[0095] While certain examples of the present disclosure refer to a processing circuitry being configured to perform the above recited operations, the functionalities / operations of the aforementioned functional modules can be performed by the one or more processors in the processing circuitry in various ways. By way of example, the operations of each module can be performed by a specific processor, or by a combination of processors. The operations of the various functional modules, such as processing the examination / inspection image, and performing defect examination, etc., can thus be performed by respective processors (or processor combinations), while, optionally, these operations may be performed by the same processor. The present disclosure should not be limited to being construed as one single processor always performing all the operations.

[0096] Those versed in the art will readily appreciate that the teachings of the presently disclosed subject matter are not bound by the particular illustrations in FIGS. 1, 2, and 4. Different components and modules in FIGS. 1, 2, and 4 can be made up of any combinationof software, hardware, and / or firmware, as relevant, executed on a suitable device or devices, which perform the functions as defined and explained herein. Equivalent and / or modified functionality, as described with respect to each component and module, can be consolidated or divided in another manner. Thus, in some examples of the presently disclosed subject matter, the system may include fewer, more, modified and / or different components, modules, and functions than those shown in FIGS. 1, 2, and 4.

[0097] Different components in FIGS. 1, 2, and 4 may represent a plurality of the particular components, which are adapted to independently and / or cooperatively operate to process various data and electrical inputs, and for enabling operations related to a computerized examination system. In some cases, multiple instances of a component may be utilized for reasons of performance, redundancy, and / or availability. Similarly, in some cases, multiple instances of a component may be utilized for reasons of functionality or application. For example, different portions of the particular functionality may be placed in different instances of the component. Any reference made in the specification to a single processing circuitry should be interpreted to optionally include multiple processing circuitries.

[0098] Unless specifically stated otherwise, as apparent from the above discussions, it is appreciated that, throughout the specification, discussions utilizing terms such as "detecting", "monitoring", "activating", "receiving", "providing" or the like, include an action and / or processes of a computer that manipulate and / or transform data into other data, said data represented as physical quantities, e.g. such as electronic quantities, and / or said data representing the physical objects.

[0099] The terms "system", "computer system" or the like, used herein, should be expansively construed to include any kind of hardware-based electronic device with one or more data processing circuitries. Each processing circuitry can comprise, for example, one or more processors operatively connected to computer memory, capable of executing stored instructions to perform the operations described herein. Any reference made in the description or claims to a processing circuitry should be construed to include alsomultiple processing circuitries.

[0100] The one or more processors referred to herein can represent one or more general-purpose processing devices, such as a microprocessor, a central processing unit, or the like. More particularly, a given processor may be one of a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a processor implementing other instruction sets, or a processor implementing a combination of instruction sets. The one or more processors may also be one or more special-purpose processing devices such as an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a digital signal processor (DSP), a graphics processing unit (GPU), a network processor, or the like.

[0101] It is appreciated that certain features of the presently disclosed subject matter, which are, for clarity, described in the context of separate embodiments, may also be provided in combination in a single embodiment. Conversely, various features of the presently disclosed subject matter, which are, for brevity, described in the context of a single embodiment, may also be provided separately, or in any suitable sub-combination.

[0102] In embodiments of the presently disclosed subject matter, fewer, more and / or different stages than those shown in FIGs. 3 and 5 can be executed. In embodiments of the presently disclosed subject matter, one or more stages, illustrated in the figures, may be executed in a different order, and / or one or more groups of stages may be executed simultaneously.

[0103] It will also be understood that the system according to the presently disclosed subject matter may be a suitably programmed computer. Likewise, the presently disclosed subject matter contemplates a computer program being readable by a computer for executing the method of the presently disclosed subject matter. The presently disclosed subject matter further contemplates a machine-readable (e.g., non-transitory) memory tangibly embodying a program of instructions executable by the machine for executing the method of the presently disclosed subject matter.It is to be understood that the presently disclosed subject matter is not limited in its application to the details set forth in the description contained herein or illustrated in the drawings. The presently disclosed subject matter is capable of other embodiments and of being practiced and carried out in various ways. Hence, it is to be understood that the phraseology and terminology employed herein are for the purpose of description and should not be regarded as limiting. As such, those skilled in the art will appreciate that the conception upon which this disclosure is based may readily be utilized as a basis for designing other structures, methods, and systems for carrying out the several purposes of the present presently disclosed subject matter.

Claims

CLAIMS:

1. An IMU (Inertial Measurement Unit) error mitigation circuitry operatively connectable to an IMU mountable onboard a platform; wherein the IMU comprises an accelerometer subsystem configured to measure linear acceleration along three orthogonal axes (x, y, z) and a rate gyro subsystem configured to measure angular velocity around three orthogonal axes (x, y, z); the accelerometer subsystem and the rate gyro subsystem being configured to provide real-time IMU data;the IMU error-mitigation circuitry is configured to:monitor the real-time IMU data to detect an IMU fault in the real-time IMU data; responsive to detecting an IMU fault in at least one IMU axis, apply a machine learning model trained to process the real-time IMU data, including accelerometer and rate gyro measurements, and reconstruct missing or degraded data resulting from the IMU fault to obtain reconstructed IMU data;wherein the machine learning model is trained to reconstruct missing or faulty IMU data for the at least one IMU axis based on partially available IMU data by leveraging learned relationships between IMU measurements, thereby generating complete IMU data that reflects the dynamic movement of the platform; andprovide reconstructed IMU data to a client.

2. The IMU error-mitigation circuitry of claim 1, wherein the machine learning model is trained using a training dataset comprising IMU data series, each IMU data series being a time-series of IMU measurements gathered over a respective trajectory.

3. The IMU error-mitigation circuitry of claim 2, wherein the IMU data series in the training dataset corresponds to platforms of a type matching that of the platform on which the IMU is mounted, ensuring that the machine learning model can address distinct motion characteristics inherent to the platform.

4. The IMU error-mitigation circuitry of any one of the preceding claims, configured to provide the reconstructed IMU data to a navigation computer onboard the platform to facilitate navigation of the platform.

5. The IMU error-mitigation circuitry of any one of the preceding claims, wherein the platform is any one of: ground vehicle, aerospace vehicle, and a marine vehicle.

6. The IMU error-mitigation circuitry of any one of the preceding claims, wherein the platform is an aerospace vehicle, and the processing circuitry is configured to generate the reconstructed IMU data during flight to enable continuous motion tracking of the platform in the presence of an IMU failure.

7. The IMU error-mitigation circuitry of any one of the preceding claims, wherein the machine learning model is a Deep Neural Network.

8. The IMU error-mitigation circuitry of any one of the preceding claims, wherein the IMU failure is a failure of a rate gyro sensor in the rate gyro subsystem configured to measure angular velocity around a respective axis.

9. The IMU error-mitigation circuitry of any one of the preceding claims, wherein the IMU failure is a failure of an accelerometer sensor in the accelerometer subsystem configured to measure linear acceleration along a respective axis.

10. The IMU error-mitigation circuitry of any one of the preceding claims, wherein the IMU data processed by the machine learning model includes, in addition to the accelerometer and rate gyro measurements, auxiliary data generated by another sensor or device onboard the platform.

11. The IMU error-mitigation circuitry of claim 10, further comprising a magnetometer configured to measure the magnetic field in three axes, wherein the auxiliary data includes magnetic field measurements obtained by the magnetometer.

12. The IMU error-mitigation circuitry of any one of the preceding claims, wherein activation of the machine learning model includes applying the machine learning model on IMU output obtained from only viable IMU sensors.

13. The IMU error-mitigation circuitry of any one of the preceding claims, wherein the processing circuitry is configured to apply a lever arm correction to the accelerometer and rate gyro measurements before applying the machine learning model, thereby enabling the application of the same machine learning model for reconstructing IMU data across different mounting placements of the IMU on the platform.

14. The IMU error-mitigation circuitry of any one of the preceding claims is integrated within the processing circuitry of the IMU.

15. A method of real-time fault-mitigation in an Inertial Measurement Unit (IMU) mounted on a platform, the method can be executed by a processing circuitry operatively connected to the IMU; the method comprising:obtaining real-time IMU data, comprising:obtaining linear acceleration measured along three orthogonal axes (x, y, z) using an accelerometer subsystem;obtaining angular velocity measured around three orthogonal axes (x, y, z) using a rate gyro subsystem;monitoring the real-time IMU data;detecting an IMU fault in at least one IMU axis of the real-time IMU data; activating a machine learning model trained to process the real-time IMU data, including accelerometer and rate gyro measurements, and reconstructing missing or degraded data resulting from the IMU fault; wherein the machine learning model is trained to reconstruct missing or faulty IMU data for at least one IMU axis based on partially available IMU data by leveraging learned relationships between different IMUmeasurements and thereby generating complete IMU data that reflects dynamics of the platform; and providing reconstructed IMU data to a client.

16. The method of claim 15, wherein the machine learning model is trained using a training dataset comprising IMU data series, each IMU data series being a timeseries of IMU measurements gathered over a respective trajectory.

17. The method of claim 16, wherein the IMU data series in the training dataset corresponds to platforms of a type matching that of the platform on which the IMU is mounted, ensuring that the machine learning model can address distinct motion characteristics inherent to the platform.

18. The method of any one of claims 15 to 17, wherein the client is a navigation computer onboard the platform, the method comprising providing the reconstructed IMU data to the navigation computer to facilitate navigating of the platform.

19. The method of any one of claims 15 to 18, wherein the platform is any one of: ground vehicle, aerospace vehicle, and a marine vehicle.

20. The method of any one of claims 15 to 19, wherein the platform is an aerospace vehicle, the method comprising generating reconstructed IMU data during flight to enable continuous motion tracking of the platform in the presence of an IMU failure.

21. The method of any one of claims 15 to 20, wherein the machine learning model is a Deep Neural Network.

22. The method of any one of claims 15 to 21, wherein the IMU failure is a failure of a rate gyro sensor in the rate gyro subsystem configured to measure angular velocity around a respective axis.

23. The method of any one of claims 15 to 22, wherein the IMU failure is a failure of an accelerometer sensor in the accelerometer subsystem configured to measure linear acceleration along a respective axis.

24. The method of any one of claims 15 to 23, wherein IMU data that is obtained and monitored includes, in addition to the accelerometer measurements and the rate gyro measurements, auxiliary data generated by another sensor or device onboard the platform.

25. The method of claim 24, further comprising measuring a magnetic field in three axes using a magnetometer, wherein the auxiliary data includes magnetic field measurements obtained by the magnetometer.

26. The method of any one of claims 15 to 25, wherein activating the machine learning model includes applying the machine learning model on IMU output obtained from only viable IMU sensors.

27. The method of any one of claims 15 to 26 further comprising applying a lever arm correction to the accelerometer and rate gyro measurements before applying the machine learning model, enabling the same machine learning model to reconstruct IMU data across different mounting placements of the IMU on the platform.

28. A non-transitory computer-readable medium comprising instructions that, when executed by a computer, cause the computer to perform a method of real-time fault-mitigation in an Inertial Measurement Unit (IMU) mounted on a platform, the method comprising:obtaining real-time IMU data, the real-time IMU data comprising:linear acceleration measurements along three orthogonal axes (x, y, z) received from an accelerometer subsystem; andangular velocity measurements around three orthogonal axes (x, y, z) received from a rate gyro subsystem;monitoring the real-time IMU data to detect an IMU fault in at least one IMU axis; detecting an IMU fault in at least one IMU axis of the real-time IMU data;responsive to detecting the IMU fault, activating a machine learning model trained to process the real-time IMU data, including accelerometer and rate gyro measurements, and reconstructing missing or degraded data resulting from the IMU fault;wherein the machine learning model is trained to reconstruct missing or faulty IMU data for at least one IMU axis based on partially available IMU data by leveraging learned relationships between different IMU measurements and thereby generating complete IMU data that reflects the dynamics of the platform; and providing reconstructed IMU data to a client.

29. An INS comprising the IMU error-mitigation circuitry according to any one of claims 1 to 14.

30. A projectile comprising the IMU error-mitigation circuitry according to any one of claims 1 to 14.

31. An IMU (Inertial Measurement Unit) with real-time IMU fault mitigation capabilities mountable on a platform; the IMU comprising:an internally integrated processing circuitry, an accelerometer subsystem, and a rate gyro subsystem;the accelerometer subsystem is configured to measure linear acceleration along three orthogonal axes (x, y, z) and the rate gyro subsystem is configured to measure angular velocity around three orthogonal axes (x, y, z); the accelerometer subsystem and the rate gyro subsystem being configured to provide real-time IMU data;the processing circuitry is configured to:monitor the real-time IMU data to detect an IMU fault in the real-time IMU data; responsive to detecting an IMU fault in at least one IMU axis, apply a machine learning model trained to process the real-time IMU data, including accelerometer and rate gyro measurements, and reconstruct missing or degraded data resulting from the IMU fault to obtain reconstructed IMU data;wherein the machine learning model is trained to reconstruct missing or faulty IMU data for the at least one IMU axis based on partially available IMU data by leveraging learned relationships between IMU measurements, thereby generating complete IMU data that reflects the dynamic movement of the platform; andprovide reconstructed IMU data to a client.

Images