Methods and systems for identifying potential vehicle defects
The inventors have developed technology to facilitate the inspection of vehicles, such as cars, for the presence of defects. The technology may be used to facilitate detection of any defects of one or more vehicle defects before, during, and/or after inspection of the vehicle. The technology includes software and trained machine learning models for performing analyses to determine likely vehicle defects prior to completion of a vehicle inspection, to determine whether any vehicle defects are present based on data acquired during the vehicle inspection, and/or to determine, after completion of the vehicle inspection, whether there are any discrepancies between any defects identified by an inspector during the vehicle inspection and defects automatically detected by analyzing data collected during the inspection of the vehicle.
1 . A method for using trained machine learning (ML) models to process data about a vehicle in furtherance of identifying one or more vehicle defects, the method comprising:
using at least one computer hardware processor to perform, subsequent to completion of an inspector's inspection of the vehicle:
obtaining an initial vehicle condition report comprising inspection results, the inspection results indicating a first set of zero, one or more vehicle defects that the inspector determined to be present in the vehicle;
obtaining data about the vehicle gathered at least in part during inspection of the vehicle by the inspector;
applying the trained ML models to the data about the vehicle to obtain a second set of zero, one or more vehicle defects that are indicated by the trained ML models to be present in the vehicle;
generating an indication of agreement or discrepancy between the first set of vehicle defects that the inspector determined to be present in the vehicle and the second set of vehicle defects that are indicated by the trained ML models to be present in the vehicle; and
outputting the initial vehicle condition report and/or the indication of agreement or discrepancy.
2 . The method of claim 1 , wherein generating the indication of agreement or discrepancy between the first set of vehicle defects and the second set of vehicle defects comprises:
processing the first set of vehicle defects and the second set of vehicle defects using a trained ML risk model to obtain a likelihood of a discrepancy being present between the first set of vehicle defects and the second set of vehicle defects.
3 . The method of claim 2 , wherein the trained ML risk model comprises a trained random forest model trained to determine whether the likelihood of the discrepancy being present exceeds a threshold.
4 . The method of claim 1 , wherein the trained ML risk model is trained to determine a likelihood a vehicle defect is present in the vehicle, the vehicle defect being from the group consisting of: engine noise, exhaust system, and structural rust.
5 . The method of claim 1 , wherein generating the indication of agreement or discrepancy between the first set of vehicle defects and the second set of vehicle defects is performed using one or more ML risk models or is performed using one or more rules and generating the indication of agreement or discrepancy involves processing the inspection results and likelihoods, obtained by applying the trained ML models to the data about the vehicle, with the one or more ML risk models or one or more rules.
6 . The method of claim 1 , wherein each of the trained ML models is trained to generate a likelihood that a respective type of a vehicle defect is present in the vehicle.
7 . The method of claim 1 , wherein the trained ML models include a first trained ML model trained to generate a first likelihood that a first type of vehicle defect is present in the vehicle, wherein the first type of vehicle defect is an undercarriage defect, engine audio defect, dashboard defect, and/or exterior/mechanical defect.
8 . The method of claim 1 , further comprising generating a revised vehicle condition report, the generating comprising:
providing the initial vehicle condition report and the indication of agreement or discrepancy between the first set of vehicle defects and the second set of vehicle defects to a reviewer through a device used by the reviewer;
receiving input from the device whether or not to modify the initial vehicle condition report based on the indication; and
generating a final vehicle condition report, based on the input received from the device.
9 . The method of claim 8 , wherein:
upon receiving input from the device to modify the initial vehicle report, the generating the final vehicle condition report comprises modifying the inspection results indicating the first set of zero, one or more vehicle defects by replacing an indication of one or more vehicle defects in the first set with one or more corresponding indications from the second set.
10 . The method of claim 8 , further comprising:
prior to receiving the input whether or not to modify the initial vehicle condition report,
providing, to an inspector, a recommendation to obtain additional inspection results; and
subsequent to obtaining additional inspection results, determining whether or not to modify the initial vehicle condition report.
11 . The method of claim 1 , wherein:
the first set of vehicle defects does not indicate a vehicle defect of a first type being present in the vehicle;
the second set of vehicle defects indicates a likelihood of the vehicle defect of the first type being present in the vehicle; and
generating the indication of agreement or discrepancy comprises generating an indication of discrepancy with respect to the vehicle defect of the first type when the likelihood exceeds a threshold.
12 . The method of claim 11 , wherein:
the trained ML models include a first trained ML model trained to detect vehicle defects of the first type, and
applying the trained (ML) models to the data about the vehicle comprises applying the first trained ML model to at least some of the data about the vehicle to obtain the likelihood of the vehicle defect of the first type being present in the vehicle.
13 . The method of claim 11 , wherein:
the first set of vehicle defects does not indicate presence of an engine vehicle defect;
the second set of vehicle defects does indicate a likelihood of the engine vehicle defect being present in the vehicle; and
generating the indication of agreement or discrepancy comprises generating an indication of discrepancy with respect to the engine vehicle defect based on the likelihood.
14 . The method of claim 11 , wherein the first trained ML model comprises:
a first neural network portion comprising a plurality of one-dimensional (1D) convolutional layers configured to process an audio waveform;
a second neural network portion comprising a plurality of two-dimensional (2D) convolutional layers configured to process a 2D representation of the audio waveform; and
a fusion neural network portion comprising one or more fully connected layers configured to combine outputs produced by the first neural network portion and the second neural network portion to obtain a likelihood indicative of the presence or absence of the at least one vehicle defect.
15 . The method of claim 11 , wherein the data about the vehicle comprises data gathered during the inspection of the vehicle using at least one hardware sensor configured to record audio of the vehicle during operation and/or capture one or more images and/or videos of the vehicle.
16 . The method of claim 11 , wherein the data about the vehicle comprises: an audio recording of the vehicle during its operation, an image of at least a portion of the vehicle, a video of at least a portion of the vehicle, and/or one or more on-board diagnostic (OBD) codes.
17 . A system, comprising:
at least one hardware processor; and
at least one non-transitory computer-readable storage medium storing processor executable instructions that when executed by the at least one computer hardware processor perform a method for using trained machine learning (ML) models to process data about a vehicle in furtherance of identifying one or more vehicle defects, the method comprising:
obtaining an initial vehicle condition report comprising inspection results, the inspection results indicating a first set of zero, one or more vehicle defects that the inspector determined to be present in the vehicle;
obtaining data about the vehicle gathered at least in part during inspection of the vehicle by the inspector;
applying the trained ML models to the data about the vehicle to obtain a second set of zero, one or more vehicle defects that are indicated by the trained ML models to be present in the vehicle;
generating an indication of agreement or discrepancy between the first set of vehicle defects that the inspector determined to be present in the vehicle and the second set of vehicle defects that are indicated by the trained ML models to be present in the vehicle; and
outputting the initial vehicle condition report and the indication of agreement or discrepancy.
18 . The system of claim 17 , wherein generating the indication of agreement or discrepancy between the first set of vehicle defects and the second set of vehicle defects comprises comparing likelihoods, generated by the trained ML models, indicating that defects of the second set of vehicle defects are present in the vehicle, to the first set of vehicle defects.
19 . At least one non-transitory computer-readable storage medium storing processor executable instructions that when executed by the at least one computer hardware processor perform a method for using trained machine learning ML models to process data about a vehicle in furtherance of identifying one or more vehicle defects, the method comprising:
obtaining an initial vehicle condition report comprising inspection results, the inspection results indicating a first set of zero, one or more vehicle defects that the inspector determined to be present in the vehicle;
obtaining data about the vehicle gathered at least in part during inspection of the vehicle by the inspector;
applying the trained (ML) models to the data about the vehicle to obtain a second set of zero, one or more vehicle defects that are indicated by the trained ML models to be present in the vehicle;
generating an indication of agreement or discrepancy between the first set of vehicle defects that the inspector determined to be present in the vehicle and the second set of vehicle defects that are indicated by the trained ML models to be present in the vehicle; and
outputting the initial vehicle condition report and the indication of agreement or discrepancy.
20 . The at least one non-transitory computer-readable storage medium of claim 19 , wherein generating the indication of agreement or discrepancy between the first set of vehicle defects and the second set of vehicle defects comprises comparing likelihoods, generated by the trained ML models, indicating that defects of the second set of vehicle defects are present in the vehicle, to the first set of vehicle defects.