# US 2025/0083834 A1 · Aircraft control system failure event search

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2025/0083834 A1
- Document ID: US20250083834A1
- Status: Published application
- Kind: A1
- Application number: 18/463,188
- Filed: 2023-09-07
- Published: 2025-03-13
- Earliest claimed: 2023-09-07
- Assignees: The Boeing Company
- Applicants: The Boeing Company
- Inventors: Ítalo Romani de Oliveira; Alejandro Torres Gámiz; Thiago Toshio Matsumoto
- Web page: https://patentgrep.com/patent/US20250083834A1

## CPC

- B64D 45/00: Aircraft indicators or protectors not otherwise provided for
- B64D 2045/0085: {Devices for aircraft health monitoring, e.g. monitoring flutter or vibration}
- B64D: EQUIPMENT FOR FITTING IN OR TO AIRCRAFT; FLIGHT SUITS; PARACHUTES; ARRANGEMENT OR MOUNTING OF POWER PLANTS OR PROPULSION TRANSMISSIONS IN AIRCRAFT

## Abstract

A computing device including a processor configured to receive a specification of a search space over inputs to an aircraft control system. The processor is further configured to perform a Monte Carlo tree search (MCTS) to identify an adversarial branch of the search space. The MCTS includes, in each of a plurality of partitioning steps, computing discrete branches of the search space by computing sample values of a partitioning objective function, estimating a probability density function of the objective function, and partitioning a current-iteration portion of the search space based on the probability density function. The discrete branches each include event nodes that have corresponding estimated event probabilities. The MCTS further includes, at a multi-agent system model including supervised learning models, recomputing the estimated event probabilities and the current-iteration portion of the search space. The processor is further configured to generate and output a graphical representation of the adversarial branch.

## Claims

### Claim 1

A computing device comprising:
a processor configured to:
receive a specification of a search space over a plurality of inputs to an aircraft control system;
perform a Monte Carlo tree search (MCTS) to identify an adversarial branch of the search space, wherein performing the MCTS includes, in each of a plurality of partitioning steps:
computing a plurality of discrete branches of the search space at least in part by:
computing a plurality of sample objective function values of a partitioning objective function for different respective sample input values of the inputs included in a current-iteration portion of the search space;
estimating a probability density function of the partitioning objective function over the current-iteration portion of the search space based at least in part on the sample objective function values; and
partitioning the current-iteration portion of the search space into the discrete branches based at least in part on the probability density function, wherein:
the discrete branches each include a respective sequence of event nodes that have corresponding estimated event probabilities; and
the plurality of event probabilities include a plurality of failure event probabilities of a respective plurality of failure events;
at a multi-agent system model that includes a plurality of supervised learning models, recomputing the estimated event probabilities associated with each of a plurality of event nodes included in the discrete branches; and
recomputing the current-iteration portion of the search space based at least in part on the failure event probabilities;
generate a graphical representation of the adversarial branch; and
output the graphical representation for display at a display device.

### Claim 2 (depends on 1)

The computing device of claim 1, wherein the processor is further configured to perform the MCTS at least in part by executing one or more reinforcement learners.

### Claim 3 (depends on 2)

The computing device of claim 2, wherein the processor is further configured to train the one or more reinforcement learners at least in part by:
computing respective reward values of the one or more reinforcement learners based at least in part on the respective failure event probabilities estimated for the failure events in the discrete branches; and
modifying respective model weights of the one or more reinforcement learners based at least in part on the reward values.

### Claim 4 (depends on 1)

The computing device of claim 1, wherein the processor is further configured to:
receive user feedback via a user-controlled agent included in the multi-agent system model; and
update the plurality of discrete branches based at least in part on the user feedback.

### Claim 5 (depends on 4)

The computing device of claim 4, wherein the user feedback includes simulated aircraft control instructions received from a flight simulator application program in response to user inputs and aircraft control logic.

### Claim 6 (depends on 5)

The computing device of claim 5, wherein the graphical representation includes an aircraft operator training scenario presented via the flight simulator application program.

### Claim 7 (depends on 4)

The computing device of claim 4, wherein the user feedback includes aircraft control system log data received from the aircraft control system.

### Claim 8 (depends on 4)

The computing device of claim 4, wherein the processor receives the user feedback as training data with which the one or more supervised learning models are trained.

### Claim 9 (depends on 1)

The computing device of claim 1, wherein the plurality of supervised learning models include an imitation learner trained to imitate behavior of an operator of the aircraft control system.

### Claim 10 (depends on 1)

The computing device of claim 1, wherein computing the plurality of discrete branches further includes discretizing one or more of the plurality of inputs to the aircraft control system.

### Claim 11 (depends on 1)

The computing device of claim 1, wherein the adversarial branch has a highest failure event probability among the plurality of discrete branches computed in the plurality of partitioning steps.

### Claim 12 (depends on 1)

The computing device of claim 1, wherein the plurality of failure events are each selected from the group consisting of:
a collision between two aircraft;
a collision between an aircraft and ground;
an aircraft component malfunction;
an aircraft destabilization;
loss of control over the aircraft;
aircraft stalling;
a hard landing; and
a runway excursion.

### Claim 13

A method for use with a computing device, the method comprising:
receiving a specification of a search space over a plurality of inputs to an aircraft control system;
performing a Monte Carlo tree search (MCTS) to identify an adversarial branch of the search space, wherein performing the MCTS includes, in each of a plurality of partitioning steps:
computing a plurality of discrete branches of the search space at least in part by:
computing a plurality of sample objective function values of a partitioning objective function for different respective sample input values of the inputs included in a current-iteration portion of the search space;
estimating a probability density function of the partitioning objective function over the current-iteration portion of the search space based at least in part on the sample objective function values; and
partitioning the current-iteration portion of the search space into the discrete branches based at least in part on the probability density function, wherein:
the discrete branches each include a respective sequence of event nodes that have corresponding estimated event probabilities; and
the plurality of event probabilities include a plurality of failure event probabilities of a respective plurality of failure events;
at a multi-agent system model that includes a plurality of supervised learning models, recomputing the estimated event probabilities associated with each of a plurality of event nodes included in the discrete branches; and
recomputing the current-iteration portion of the search space based at least in part on the failure event probabilities;
generating a graphical representation of the adversarial branch; and
outputting the graphical representation for display at a display device.

### Claim 14 (depends on 13)

The method of claim 13, wherein performing the MCTS includes executing one or more reinforcement learners.

### Claim 15 (depends on 14)

The method of claim 14, further comprising training the one or more reinforcement learners at least in part by:
computing respective reward values of the one or more reinforcement learners based at least in part on the respective failure event probabilities estimated for the failure events in the discrete branches; and
modifying respective model weights of the one or more reinforcement learners based at least in part on the reward values.

### Claim 16 (depends on 13)

The method of claim 13, further comprising:
receiving user feedback via a user-controlled agent included in the multi-agent system model; and
updating the plurality of discrete branches based at least in part on the user feedback.

### Claim 17 (depends on 16)

The method of claim 16, wherein the user feedback includes simulated aircraft control instructions received from a flight simulator application program in response to user inputs and aircraft control logic.

### Claim 18 (depends on 16)

The method of claim 16, wherein the user feedback includes aircraft control system log data received from the aircraft control system.

### Claim 19 (depends on 13)

The method of claim 13, wherein the plurality of failure events are each selected from the group consisting of:
a collision between two aircraft;
a collision between an aircraft and ground;
an aircraft component malfunction;
an aircraft destabilization;
loss of control over the aircraft;
aircraft stalling;
a hard landing; and
a runway excursion.

### Claim 20

A computing device comprising:
a processor configured to:
receive a specification of a search space over a plurality of inputs;
discretize one or more of the plurality of inputs;
perform a Monte Carlo tree search (MCTS) to identify an adversarial branch of the search space, wherein:
the adversarial branch has a highest failure event probability among the plurality of discrete branches computed in a plurality of partitioning steps; and
performing the MCTS includes, in each of the plurality of partitioning steps:
computing a plurality of discrete branches of the search space at least in part by:
computing a plurality of sample objective function values of a partitioning objective function for different respective sample input values of the inputs included in a current-iteration portion of the search space;
estimating a probability density function of the partitioning objective function over the current-iteration portion of the search space based at least in part on the sample objective function values; and
partitioning the current-iteration portion of the search space into the discrete branches based at least in part on the probability density function, wherein:
the discrete branches each include a respective sequence of event nodes that have corresponding estimated event probabilities; and
the plurality of event probabilities include a plurality of failure event probabilities of a respective plurality of failure events;
at a multi-agent system model that includes a plurality of supervised learning models, recomputing the estimated event probabilities associated with each of a plurality of event nodes included in the discrete branches; and
recomputing the current-iteration portion of the search space based at least in part on the failure event probabilities; and
output an indication of the adversarial branch.
