Abstract: A data generation unit generates second abnormal sound data indicating a change in loudness of a sound each time a rotation state changes by a certain amount, based on first abnormal sound data that is data of an abnormal sound generated due to a rotary motion of a rotation device mounted on an object, and rotation data that is data relating to the rotary motion of the rotation device. The second abnormal sound data is input to a sound source estimation unit, and a certainty factor of being a sound source of the abnormal sound indicated by the second abnormal sound data is output to each of a plurality of parts mounted on a vehicle.

Abstract: A misfire detection device for an internal combustion engine is provided. A mapping takes time series data of instantaneous speed parameters as inputs. Each instantaneous speed parameter corresponds to one of a plurality of successive second intervals in a first interval. The instantaneous speed parameters correspond to the rotational speed of the crankshaft. The first interval is a rotational angular interval of the crankshaft in which compression top dead center occurs. The second interval is smaller than an interval between compression top dead center positions. The mapping outputs a probability that a misfire has occurred in at least one cylinder that reaches compression top dead center in the first interval. The mapping data defining the mapping has been learned by machine learning.

Abstract: A data generation unit generates second abnormal sound data indicating a change in loudness of a sound each time a rotation state changes by a certain amount, based on first abnormal sound data that is data of an abnormal sound generated due to a rotary motion of a rotation device mounted on an object, and rotation data that is data relating to the rotary motion of the rotation device. The second abnormal sound data is input to a sound source estimation unit, and a certainty factor of being a sound source of the abnormal sound indicated by the second abnormal sound data is output to each of a plurality of parts mounted on a vehicle.

Abstract: A CPU calculates probability models by inputting input variables to a neural network. Next, the CPU determines whether a maximum value among the calculated probability models is larger than an upper limit value of a permissible range. When the maximum value is larger than the upper limit value of the permissible range, the CPU calculates a difference between the maximum value and a first reference value within the permissible range, and subtracts the difference from all the probability models. Then, the CPU calculates a probability of misfire in each cylinder by inputting each of the calculated probability models to a softmax function of mapping.

Abstract: A misfire detection device for an internal combustion engine is provided. A mapping takes time series data of instantaneous speed parameters as inputs. Each instantaneous speed parameter corresponds to one of a plurality of successive second intervals in a first interval. The instantaneous speed parameters correspond to the rotational speed of the crankshaft. The first interval is a rotational angular interval of the crankshaft in which compression top dead center occurs. The second interval is smaller than an interval between compression top dead center positions. The mapping outputs a probability that a misfire has occurred in at least one cylinder that reaches compression top dead center in the first interval. The mapping data defining the mapping has been learned by machine learning.

Abstract: A misfire detection device for an internal combustion engine is provided. A mapping takes time series data of instantaneous speed parameters as inputs. Each instantaneous speed parameter corresponds to one of a plurality of successive second intervals in a first interval. The instantaneous speed parameters correspond to the rotational speed of the crankshaft. The first interval is a rotational angular interval of the crankshaft in which compression top dead center occurs. The second interval is smaller than an interval between compression top dead center positions. The mapping outputs a probability that a misfire has occurred in at least one cylinder that reaches compression top dead center in the first interval. The mapping data defining the mapping has been learned by machine learning.

Abstract: A CPU calculates probability models by inputting input variables to a neural network. Next, the CPU determines whether a maximum value among the calculated probability models is larger than an upper limit value of a permissible range. When the maximum value is larger than the upper limit value of the permissible range, the CPU calculates a difference between the maximum value and a first reference value within the permissible range, and subtracts the difference from all the probability models. Then, the CPU calculates a probability of misfire in each cylinder by inputting each of the calculated probability models to a softmax function of mapping.

Abstract: A catalyst state estimation apparatus includes a first sensor, a memory and a processor. The first sensor is configured to acquire information about a catalyst that removes a toxic substance in an exhaust gas, the first sensor being provided in a main passage into which the exhaust gas flows from an internal combustion engine. The memory is configured to previously store a catalyst state estimation model that includes at least one mathematical model. The processor is configured to estimate a removal performance of the catalyst by applying the information about the catalyst acquired by the first sensor to the catalyst state estimation model.

Abstract: A misfire detection device for an internal combustion engine is provided. A mapping takes time series data of instantaneous speed parameters as inputs. Each instantaneous speed parameter corresponds to one of a plurality of successive second intervals in a first interval. The instantaneous speed parameters correspond to the rotational speed of the crankshaft. The first interval is a rotational angular interval of the crankshaft in which compression top dead center occurs. The second interval is smaller than an interval between compression top dead center positions. The mapping outputs a probability that a misfire has occurred in at least one cylinder that reaches compression top dead center in the first interval. The mapping data defining the mapping has been learned by machine learning.

Abstract: A catalyst state estimation apparatus includes a first sensor, a memory and a processor. The first sensor is configured to acquire information about a catalyst that removes a toxic substance in an exhaust gas, the first sensor being provided in a main passage into which the exhaust gas flows from an internal combustion engine. The memory is configured to previously store a catalyst state estimation model that includes at least one mathematical model. The processor is configured to estimate a removal performance of the catalyst by applying the information about the catalyst acquired by the first sensor to the catalyst state estimation model.