JPH09315261A - Sum-of-product arithmetic circuit learning method and device thereof as well as starting controller of occupant protective device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To surely enable discrimination whether an occupant protecting device should be started or not with certainty irrespective of the form of a car crash. SOLUTION: A G sensor 41 detects the extent of deceleration to be added to a vehicle 32, outputting this deceleration as a G signal. An interval integrating circuit 43 performs an interval integration of 10ms duration to this G signal, outputting the interval integral signal. A serial-parallel converter circuit 45 inputs those of interval integral signals or time series data successively in series and thereby it performs a serial-parallel conversion, outputting these signals in parallel. A sum-of-product arithmetic circuit 46 inputs the interval integral signals from an input terminal in parallel, and it performs the operation of sum of products used with plural operational constants to these signals, outputting these signals from each output terminal. Signals out of the sum-of-product arithmetic circuit 46 are outputted from an AND circuit 64 as a starting control signal of an air bag unit 50 after logical operation is carried out in an OR circuit 66 and this AND circuit 64. These plural operational constants in the sum-of-product arithmetic circuit 46 are determined by the learning used with a neural network.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an occupant protection device such as an airbag device for protecting an occupant in a vehicle when the vehicle collides, and more particularly to controlling activation of such an occupant protection device. The present invention relates to a start-up control device, a product-sum operation circuit learning method and a device for determining a plurality of operation constants in a product-sum operation circuit used in the start-up control device by learning.

[0002]

2. Description of the Related Art As a device for controlling activation of an occupant protection device, there is, for example, a device for controlling ignition of a squib in an airbag device. In an airbag device, a gas generating agent is ignited by a squib in an inflator to generate gas from the inflator, and the bag is inflated by the gas to protect an occupant from collision with indoor parts.

A device for controlling ignition of a squib of such an airbag device is usually provided with a deceleration sensor (so-called G sensor) for detecting a deceleration applied to the vehicle (in particular, a deceleration in the longitudinal direction of the vehicle). The ignition control is performed based on the deceleration detection signal obtained by the G sensor. For example, the signal value of the deceleration detection signal or its integrated value is compared with an appropriate threshold value, and when the signal value or the integrated value exceeds the threshold value, control is performed so that ignition is performed by the squib.

However, when the activation of the occupant protection device is controlled by comparing the signal value or the integrated value of the deceleration detection signal with the threshold value in this way, the collision mode of the vehicle (that is, the mode of collision, the vehicle speed at the time of collision). Depending on the collision target such as the collision partner), it is necessary to perform a complicated calculation to surely determine whether or not the occupant protection device should be activated. For example,
If the collision mode of the vehicle is other than the normal collision (for example, oblique collision, offset collision, underride collision or pole collision) and the vehicle speed at the time of collision is relatively high, the occupant protection device is activated. However, when the collision is a direct collision and the vehicle speed at the time of the collision is relatively slow, it is not enough to activate the occupant protection device. However, comparing deceleration detection signals obtained from the G sensor in both cases, a predetermined time (request time) from the occurrence of a vehicle collision
Since there is not much difference between the two, as described above, it is necessary to distinguish between the two by complicated calculation, and there is a problem that the cost for constructing a calculation system or the like is high. For example, in the technique disclosed in Japanese Patent Laid-Open No. 6-92199, displacement data (a value obtained by integrating the acceleration signal twice) obtained by an acceleration signal is input to a neural network, and based on the output of this neural network, an occupant protection device It controls startup.

[0005]

However, in the learning method of the neural network of the prior art, although the waveform data of various collision modes are used as the learning signal, the occupant protection device is recognized by recognizing each collision mode. It does not learn whether to activate. Therefore, even if the calculation constant (weighting coefficient) obtained as the learning result is such that the output signal (passenger protection device operation or non-operation signal) as requested can be obtained, this calculation constant is not always required. There is no guarantee that the optimum value will be obtained by learning the neural network. Therefore, when this neural network is used as an activation identification device for an occupant protection device, there is a problem that it must be used in combination with other calculations.

Therefore, an object of the present invention is to solve the above-mentioned problems of the prior art and to reliably determine whether or not the occupant protection device should be activated regardless of the collision mode of the vehicle. The present invention is to provide a start-up control device.

[0007]

Means for Solving the Problem and Actions and Effects Thereof In order to achieve at least a part of the above-mentioned object, the first invention is a signal obtained from a vehicle impact signal corresponding to the magnitude of the impact applied to the vehicle. Is input, the product-sum operation using a plurality of operation constants is performed on the signal, and a signal for obtaining a start-up control signal for controlling the start-up of the occupant protection device is output. A method for learning a product-sum operation circuit for determining a plurality of operation constants for a circuit by learning, comprising: (a) a plurality of product-sum operation elements, a plurality of input terminals and a plurality of output terminals; A step of preparing a learning product-sum operation circuit capable of adjusting the operation constants of the plurality of product-sum operation elements; and (b) a series of vehicle-impact signals for learning, which are time-series signals prepared in advance, in series-parallel. The step of converting, (c The learning-vehicle impact signal that has been subjected to the serial-parallel conversion is input in parallel to each input end of the learning product-sum operation circuit, and the operation constants are respectively calculated by the plurality of product-sum operation elements in the learning product-sum operation circuit. A step of performing the sum-of-products calculation used, (d) a plurality of output signals output from the respective output terminals of the above-mentioned sum-of-products for arithmetic calculation circuit as the calculation result, and the vehicle impact signal for learning prepared in advance A plurality of ideal output signals, so that each error of each has a predetermined value, the step of adjusting the operation constants of the plurality of product-sum operation elements in the learning product-sum operation circuit,
The gist is to determine the operation constants of the plurality of product-sum operation elements obtained as a result of the adjustment as the plurality of operation constants in the product-sum operation circuit for the start control device.

In the first aspect of the invention, first, the learning vehicle impact signal, which is a time-series signal, is serial-parallel converted. Next, the learning-based vehicle impact signals that have undergone the series-parallel conversion are input in parallel to the respective input terminals of the learning product-sum operation circuit, and the product-sum operation is performed by using a plurality of product-sum operation elements, respectively.
As a result of the calculation, each error between the plurality of output signals output from each output terminal of the learning product-sum calculation circuit and the plurality of ideal output signals corresponding to the vehicle impact signal for learning has a predetermined value. Then, the operation constants of the plurality of product-sum operation elements in the learning product-sum operation circuit are adjusted so that the learning product-sum operation circuit learns. The arithmetic constants thus obtained by learning are determined as a plurality of arithmetic constants in the product-sum arithmetic circuit for the start control device.

Therefore, according to the first aspect of the present invention, the learning vehicle impact signal, which is a time-series signal, is subjected to serial-parallel conversion and input to the product-sum calculation circuit, so that the waveform of the learning vehicle impact signal is taken into consideration. Can be learned by the learning product-sum calculation circuit, and as the ideal output signal, a plurality of signals corresponding to the learning vehicle impact signal, in other words, a plurality of signals corresponding to the collision mode of the vehicle are prepared, By learning the product-sum calculation circuit for learning, it is possible to perform learning in a form that also reflects the collision form of the vehicle. Therefore, finally, it is possible to obtain a plurality of operation constants in the product-sum operation circuit suitable for determining whether or not to activate the occupant protection device in accordance with the collision mode of the vehicle.

In the product-sum operation circuit learning method of the first invention, it is preferable that the step (a) includes a step of preparing a neural network as the learning product-sum operation circuit.

By using a neural network as the learning product-sum operation circuit, the product-sum operation circuit can be easily learned.

In the product-sum operation circuit learning method according to the first aspect of the present invention, in the step (b), as the learning vehicle impact signals, a plurality of vehicle impact signals obtained at the time of a vehicle collision and having different collision modes from each other are obtained. In addition to the step of preparing, it is preferable that the step (d) includes the step of preparing the plurality of ideal output signals which are different for each of the plurality of vehicle impact signals.

As described above, by using a plurality of vehicle impact signals having different vehicle collision modes as learning vehicle impact signals, the learning product-sum calculation circuit is learned while sufficiently reflecting the vehicle collision mode. You can

In the product-sum operation circuit learning method according to the first aspect of the present invention, in the step (b), a vehicle impact signal other than the vehicle impact signal obtained during traveling on a rough road of the vehicle is prepared as the learning vehicle impact signal. It is preferable to include a step.

The peak waveform of the vehicle impact signal during traveling on a rough road and the peak waveform of the vehicle impact signal at the time of a high-speed head-on collision are very similar to each other and are difficult to distinguish. In addition, if a vehicle impact signal during traveling on a rough road is also used as the learning vehicle impact signal, the arithmetic constants do not easily converge. Therefore, in this way, by excluding the vehicle impact signal during traveling on a rough road and allowing the learning product-sum operation circuit to learn, the optimum value as the operation constant can be derived.

According to a second aspect of the present invention, a signal obtained from a vehicle impact signal corresponding to the magnitude of the impact applied to the vehicle is input, and a sum of products operation using a plurality of operation constants is performed on the signal to obtain an occupant. A product-sum operation circuit learning device for determining the plurality of operation constants by learning for a product-sum operation circuit for a start control device that outputs a signal for obtaining an activation control signal that controls the activation of the protection device, , A series-parallel conversion means for converting the learning vehicle impact signal, which is a time-series signal, into a series-parallel conversion, and a plurality of product-sum calculation elements, and a plurality of input ends and a plurality of output ends, The vehicle impact signal for learning that has undergone serial-parallel conversion is input in parallel from the input end, and the product-sum calculation using each of the calculation constants is performed by the plurality of product-sum calculation elements, and Adjust the calculation constant A learning product-sum operation circuit capable of performing, an output signal output from each output end of the learning product-sum operation circuit as an operation result, and a plurality of previously prepared corresponding to the learning vehicle impact signal. And an arithmetic constant adjusting means for adjusting arithmetic constants of the plurality of product-sum operation elements in the learning product-sum operation circuit so that each error between the ideal output signal and the ideal output signal becomes a predetermined value. The gist of the present invention is to determine the resulting operation constants of the product-sum operation elements as the operation constants of the product-sum operation circuit for the start control device.

According to the second aspect of the invention, the learning vehicle impact signal, which is a time-series signal, is serial-parallel converted by the serial-parallel converter. Then, the learning-use sum-of-products arithmetic circuit inputs the learning-use vehicle impact signals, which have been subjected to the series-to-parallel conversion, in parallel from the respective input terminals, and the plurality of sum-of-products calculating elements respectively perform the sum-of-products calculation using the operation constants. At this time, the error between the output signal output from each output end of the learning product-sum calculation circuit as the calculation result and the plurality of ideal output signals corresponding to the learning vehicle impact signal is calculated by the calculation constant adjusting means. A plurality of operation constants in the learning product-sum operation circuit are adjusted so that the values become predetermined values.

Therefore, according to the second aspect of the present invention, the learning product-sum operation circuit can be learned in a form that also reflects the collision form of the vehicle, and finally the collision form of the vehicle can be dealt with. It is possible to obtain a plurality of operation constants in the product-sum operation circuit suitable for determining whether or not to activate the occupant protection device. A third aspect of the present invention is an activation control device for an occupant protection device, which obtains an activation control signal for controlling activation of the occupant protection device based on a vehicle impact signal corresponding to the magnitude of an impact applied to the vehicle. A series-parallel conversion means for converting the vehicle impact signal into a series-parallel conversion, and a plurality of input ends and a plurality of output ends, the vehicle impact signal subjected to series-parallel conversion is input in parallel from the input end, A product-sum operation circuit that performs a product-sum operation using a plurality of operation constants and outputs a signal obtained as a result of the operation from each output terminal, and outputs from the signal output from the product-sum operation circuit. The gist is to obtain the activation control signal and to use the operation constants determined by the product-sum operation circuit learning method in the first invention as the plurality of operation constants in the product-sum operation circuit.

As described above, in the third aspect of the invention, first, the vehicle-impact signal, which is a time-series signal, is serial-parallel converted by the serial-parallel converter. Then, the series-parallel converted vehicle impact signal is input in parallel to a plurality of input terminals of the product-sum calculation circuit, and a product-sum calculation using a plurality of calculation constants is performed.
A signal obtained as a result of the calculation is output from each of the plurality of output terminals. Then, an activation control signal for controlling activation of the occupant protection device is obtained from those output signals. At this time, as the plurality of operation constants in the product-sum operation circuit, the operation constants determined by the product-sum operation circuit learning method according to the first aspect are used.

As described above, in the third invention, the optimum arithmetic constant learned for each collision form, which is obtained by the product-sum arithmetic circuit learning method in the first invention, is used as the arithmetic constant of the product-sum arithmetic circuit. Therefore, the product-sum operation circuit is
It is possible to output a plurality of output signals according to the collision mode. Therefore, it is possible to determine whether or not the occupant protection device should be activated in accordance with the waveform of the vehicle impact signal input to the product-sum calculation circuit, that is, the collision mode of the vehicle.

Therefore, according to the third aspect of the present invention, it is possible to reliably determine whether or not the occupant protection device should be activated regardless of the collision mode of the vehicle.

In the activation control device according to the third aspect of the invention, one of the plurality of signals output from each output terminal of the product-sum calculation circuit determines whether or not to activate the occupant protection device. It is preferable that the signal is

Further, in the start-up control device according to the third aspect of the invention, the plurality of signals output from the respective output terminals of the product-sum calculation circuit can identify the collision mode when the vehicle collides. It is preferable to include possible signals.

By including such a signal capable of identifying the collision mode, the activation of the occupant protection device can be controlled in a manner corresponding to the collision mode. For example,
If there are multiple occupant protection devices, depending on the type of collision,
It is also possible to change the occupant protection device to be activated.

In the activation control device according to the third aspect of the present invention, it is preferable that the vehicle impact signal is a signal obtained by integrating the acceleration applied to the vehicle.

By using such a signal as the vehicle impact signal, a more accurate start control signal can be obtained.

In the start-up control device according to the third aspect of the present invention, based on the vehicle impact signal, a determination means for deriving a determination signal for determining whether the vehicle is in a collision or traveling on a rough road, and the sum of products operation. When the signal output from the circuit indicates that the occupant protection device should be activated, and the determination signal derived by the determination means indicates a vehicle collision, the activation control signal It is preferable that the apparatus further comprises start control signal output means for outputting a signal for starting the passenger protection device.

As described above, by further providing the determining means for determining whether the vehicle is in a collision or traveling on a bad road, the product-sum calculation circuit is in such a vehicle collision or traveling on a bad road. It is no longer necessary to determine whether or not
Therefore, when deriving a plurality of operation constants in the product-sum operation circuit by learning, the vehicle impact signal for learning is excluded from the vehicle impact signal during traveling on a bad road, and only the vehicle impact signal at the time of a vehicle collision is used as the product impact signal for learning. Since the sum calculation circuit may be trained, the optimum value can be derived as the calculation constant.

A fourth aspect of the present invention is an activation control device for an occupant protection device, which obtains an activation control signal for controlling activation of the occupant protection device based on a vehicle impact signal corresponding to the magnitude of the impact applied to the vehicle. A serial-parallel conversion means for converting the vehicle impact signal, which is a time-series signal, into a serial-parallel conversion, and a plurality of input ends and a plurality of output ends, the serial-parallel converted vehicle impact signal being input from the input end. A vehicle collision based on the vehicle impact signal, which is input in parallel, performs a product-sum operation using a plurality of operation constants, and outputs a signal obtained as a result of the operation from each output end, and a vehicle impact signal based on the vehicle impact signal. And a determination means for deriving a determination signal for determining whether the vehicle is traveling on a rough road, and the signal output from the product-sum calculation circuit indicates that the occupant protection device should be activated, and , The above When the determination signal derived by another means indicates a vehicle collision, start control signal output means for outputting a signal for starting the occupant protection device as the start control signal, and the product Determined by the product-sum operation circuit learning method according to the first aspect of the present invention, which comprises a step of preparing a learning vehicle impact signal other than the vehicle impact signal obtained during traveling on a rough road of the vehicle as the plurality of operation constants in the sum operation circuit. The gist is to use the calculated arithmetic constants.

According to the fourth aspect of the present invention, the determining means is separately provided, and the product-sum calculation circuit does not need to determine whether the vehicle is in a collision or traveling on a rough road. As the calculation constant of, the calculation constant determined by learning by the learning vehicle impact signal excluding the vehicle impact signal during traveling on a rough road is used.

By using such operation constants, the product-sum operation circuit does not take the rough road traveling into consideration,
A vehicle collision can be determined, and a more accurate start control signal can be obtained.

The word "output end" is used in the description of the claims and the configuration of each of the above-mentioned inventions, but this means that the output end (end) of the arithmetic circuit is the same. The present invention is not limited to this, and includes those existing in the middle of the circuit such as the output layer. The same applies to the word "input end".

[0033]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below based on Examples. FIG. 1 is a block diagram showing the configuration of a start-up control device for an airbag device according to a first embodiment of the present invention.

As shown in FIG. 1, the starting control device 40 of the present embodiment mainly comprises a deceleration sensor (G sensor) 41,
An analog / digital converter (A / D converter) 42, a section integration circuit 43, a serial / parallel conversion circuit 45, a product-sum operation circuit 46, an AND circuit 64, and an OR circuit 66 are provided. Then, the activation control device 40 controls the airbag device 50 via the drive circuit 52.
Is connected to the squib 51 inside. Each of these components is mounted in a predetermined position in the vehicle 32.

First, the G sensor 41 detects the deceleration applied to the vehicle 32 in the front-rear direction and outputs the detected deceleration as an analog deceleration detection signal (G signal). The G signal thus obtained becomes a signal corresponding to the magnitude of the impact applied to the vehicle 32. The G sensor 41 may be a sensor that detects only deceleration or a sensor that detects acceleration in a broad sense including deceleration.

Next, the A / D converter 42 inputs the G signal output from the G sensor 41, converts the analog signal into a digital signal, and outputs the digital signal to the interval integration circuit 43. When the G signal output from the G sensor 41 is a digital signal, the A / D converter 42 can be omitted.

Next, the section integration circuit 43 performs section integration for 10 ms on the input G signal and outputs the section integration signal. That is, the signal value of the G signal is
Assuming that (t), the signal value (that is, the interval integrated value) v10 of the interval integrated signal can be expressed as shown in Expression (1).

[0038]

[Equation 1]

Thus, by obtaining the section integral value from the G signal, the moving speed of the occupant in the vehicle can be obtained, and by using this section integral value, the start control signal can be obtained.
More accurate activation control of the airbag device can be performed.

Next, the serial / parallel conversion circuit 45
Sequentially inputs in series the interval integration signals which are time series data output from the interval integration circuit 43, performs serial-parallel conversion, and outputs in parallel.

Next, the product-sum calculation circuit 46 is provided with a plurality of product-sum calculation elements (not shown) and has 100 input terminals and 11 output terminals. The product-sum operation circuit 46 inputs the interval integration signals serial-parallel converted by the serial-parallel conversion circuit 45 in parallel from the respective input terminals,
The product-sum operation using the operation constants is performed by each internal product-sum operation element. Then, binary signals (“0”, “1”) obtained as the result of the calculation are output from the respective output terminals.

The signal output from the product-sum calculation circuit 46 is input to the OR circuit 66 and the AND circuit 64, and after logical operation is performed, it is finally output from the AND circuit 64 as a start control signal for the airbag device 50. To be done.

In this embodiment, the product-sum calculation circuit 46 having a plurality of product-sum calculation elements is used in order to derive the activation control signal from the interval integration signal output from the serial-parallel conversion circuit 45. The operation constant of each product-sum calculation element in the product-sum calculation circuit 46 is determined by learning using a neural network as described later so that an optimum start control signal can be derived.

FIG. 2 is a block diagram showing the configuration of a product-sum calculation circuit learning device for determining the calculation constant of the product-sum calculation element in the product-sum calculation circuit 46 of FIG. 1 by learning. FIG.
As shown in FIG. 2, the product-sum calculation circuit learning device 20 includes an A / D converter 22, a section integration circuit 23, a serial / parallel conversion circuit 25, a neural network 26, and an adjustment circuit 28.

Of these, the A / D converter 22 is shown in FIG.
It has the same configuration and the same function as the A / D converter 42 shown in, and inputs a G signal for learning prepared in advance and converts the G signal from an analog signal to a digital signal. As the G signal for learning, the G signal that will be obtained at the time of a vehicle collision is not necessary to start the airbag device (when the squib ignition is on) and when the airbag device is to be activated (squib ignition). In the case of the off requirement), it is prepared separately for each collision mode and each collision object. That is, for example, a total of 10 types of collision modes are assumed by combining five modes of normal collision, oblique collision, offset collision, underride collision, and pole collision and two collision objects A and B. For each of the above, a G signal when the airbag device should be activated (for example, in the case of a collision at a medium speed or a high speed) and a G signal when it is not sufficient to be activated (for a collision at a low speed) are prepared. Then, a total of 20 G signals are prepared as learning G signals.

At this time, for example, a G signal for about 250 ms is prepared as the G signal when the airbag device is not activated (when the ignition of the squib is off).
Further, as the G signal when the airbag device is to be activated (when the squib is on for ignition), for example, a determination request time from the time of a vehicle collision (that is, for determining whether or not to activate the airbag device). The G signal that is obtained until the maximum allowable time) has elapsed is prepared. In this case, not only the G signal from the time of the vehicle collision to the time when the determination request time has elapsed, but also the G signal from the time before the vehicle collision may be included.

The interval integration circuit 23 has the same configuration and the same function as the interval integration circuit 43 shown in FIG.
The learning G signal converted by the / D converter 22 is input, the section integration is performed on the G signal for 10 ms, and the section integration signal is output. Specifically, the interval integration is performed according to the above-mentioned formula (1).

The serial / parallel conversion circuit 25 is shown in FIG.
The serial / parallel conversion circuit 45 has the same configuration and the same function as the serial / parallel conversion circuit 45 shown in FIG. 1, and serially inputs the learning interval integration signal, which is time-series data output from the interval integration circuit 23, for serial / parallel conversion. And output in parallel. Specifically, as the G signal when the airbag device is not activated (when the squib is turned off), for example, the G signal for about 250 ms is A / D.
If it is input to the converter 22, a section integration signal for about 250 ms is serially input to the serial / parallel conversion circuit 25, and the serial / parallel conversion circuit 25 outputs 100 signals at intervals of about 2.5 ms. To output in parallel. Further, as the G signal when the airbag device is to be activated (when the squib is on for ignition), for example, the G signal obtained from the time of the above-described vehicle collision until the determination required time elapses is the A / D converter. The case of being input to 22 is similar to the case of the ignition off requirement.

The neural network 26 has a plurality of neuron models and has a plurality of input ends and a plurality of output ends. The neural network 26 inputs the learning interval integral signals serial-parallel converted by the serial-parallel conversion circuit 45 in parallel from the respective input ends, in substantially the same manner as the product-sum operation circuit 46 shown in FIG. A product-sum operation using operation constants is performed by the neuron model, and binary signals (“0”, “1”) obtained as the operation result are output from the respective output terminals. However, in the neural network 26, it is possible to change and adjust the operation constant in each neuron model.

The adjusting circuit 28 receives a plurality of output signals output from the output terminals of the neural network 26, and outputs a plurality of ideal output signals prepared in advance corresponding to these output signals as teacher data. , The error between the output signal and the corresponding ideal output signal is obtained, and the operation constants of the neuron models in the neural network 26 are changed and adjusted so that these errors become substantially zero. , Let the neural network 26 learn.

FIG. 3 is a block diagram showing the internal structure of the neural network 26 shown in FIG. A neural network (neural network) is an information processing system developed by imitating a neural network having a multi-layered structure of a large number of nerve cells in the brain. As shown in FIG. 3, the neural network 26 used in this embodiment.
Is divided into 3 layers of input layer, middle layer and output layer,
Each layer is composed of a neuron model 30 that imitates a nerve cell in the brain.

FIG. 4 is an explanatory diagram for explaining the neuron model 30 shown in FIG. 3 and the sigmoid function fs (X) used in the neuron model 30. FIG.
As shown in (a), the neuron model 30 is modeled as a weighted multi-input / single-output system, and is considerably simplified as compared with an actual nerve cell in the brain.

In this neuron model 30, the input signals x1 and x are input from the other i neuron models, respectively.
2, ..., xi are received. Then, as shown in equation (2), the input signals x1, x2, ..., xi are multiplied by weighting factors w1, w2 ,. Obtain the state quantity X.

[0054]

[Equation 2]

Next, the state quantity X is converted by the transfer function f (X) as shown in the equation (3) to obtain the output signal y. y = f (X) ……… (3)

Here, the transfer function f (X) is expressed as shown in equation (4). f (X) = fs (X) + α (4) In Expression (4), fs (X) is a sigmoid function, and α is a bias value, which is one of the operation constants.

The sigmoid function fs (X) is a function as shown in the equation (5), and changes with respect to the state quantity X as shown in FIG. 4 (b). fs (X) = 1 / {1 + exp (-X)} ... (5)

In the neural network 26,
The weighting factors w1, w2, ..., Wi, which are arithmetic constants, and the bias value α can be changed within respective predetermined ranges.

In the neural network 26 shown in FIG. 3, the input layer is composed of 100 neuron models 30, and each neuron model 30 is serial-parallel converted as an input signal by the serial-parallel conversion circuit 25. The section integration signals v10 are input. In the serial / parallel conversion circuit 45, the interval integrated signal v10 which is the input time series data is input.
By extracting 100 pieces of data from the data and performing serial-parallel conversion, the interval integration signal v10 is parallelized to obtain v10 (k), v10 (k-1), v10 (k-
2), ..., Output as 100 signals of v10 (k-99). Therefore, 1 of the neural network 26
These 100 interval integration signals are input in parallel to 00 input terminals.

The intermediate layer is composed of a large number of neuron models 30 in the initial stage, but is selected by learning and finally is composed of about 150 neuron models 30.

Further, the output layer is composed of 11 neuron models 30, and each neuron model 3
From 0, a plurality of output signals corresponding to the collision mode of the vehicle are output.

In the neuron model 30 shown in FIG. 3, the neuron model having two or more output lines branch outputs the same output signal y to each output line.

On the other hand, the adjusting circuit 28, as described above,
The error between a plurality of output signals output from each output terminal of the neural network 26 and an ideal output signal prepared in advance corresponding to each of these output signals is calculated so that the error becomes zero. The calculation constants (ie, weighting factors w1, w2, ..., Wi and bias value α) of each neuron model 30 in the neural network 26 are changed and adjusted, whereby the neural network 26 is made to learn. The learning of the neural network 26 is performed by using all the prepared G for learning.
The signals are performed in sequence. In other words, there are five types of collision including head-on collision, oblique collision, offset collision, underride collision and pole collision.
One mode and each combination of two collision objects A and B, when the airbag device should be activated (for example, in the case of a collision at a medium speed or a high speed), it is not enough to be activated with a G signal ( For example, learning is performed for each G signal (in the case of a collision at a low speed). Then, when all the learning is completed, the operation constants of the neuron models 30 in the neural network 26 are respectively determined.

The ideal output signal is G for learning.
When a signal is input to the A / D converter 22, an ideal signal is prepared to be output from each output terminal of the neural network 26 corresponding to the G signal.

In the present embodiment, as shown in FIG. 3, the uppermost output end of the 11 output ends of the neural network 26 is not sufficient to activate the airbag device (squib ignition OFF). ) Is output. Then, from the top, the second and subsequent signals indicate that the collision mode is head-on collision and the collision object is A, and the collision mode is head-on collision and the collision object is B. A signal indicating that the collision mode is oblique collision and the collision object is A, a signal indicating that the collision mode is oblique collision and the collision object is B, ... It is supposed to output. Therefore, for example, when the input G signal for learning is in a collision mode, the collision object is A, and the collision object is A and does not reach the drive of the airbag device (in the case of the squib ignition OFF requirement). In the case of a signal, it is ideal that signals of (1, 1, 0, 0, 0, ..., 0) are output from each output end of the neural network 26 in order from the top. Therefore, in this case, a signal of (1, 1, 0, 0, 0, ..., 0) is prepared as the ideal output signal. Further, when the input learning G signal is a G signal when the collision mode is a diagonal collision, the collision target is B, and the airbag device should be driven (in the case where the squib is turned on). From the output end of the neural network 26 to (0, 0, 0, 0, 1, ...
Ideally, the signal 0) is output. Therefore, in this case, (0, 0, 0, 0, 1,
,, 0) signal is prepared.

In this way, the product-sum operation circuit learning device 20
The operation constant of each neuron model 30 determined by learning is used as the operation constant of each product-sum operation element in the product-sum operation circuit 46 shown in FIG.

That is, in the product-sum calculation circuit 46 shown in FIG. 1, the product-sum calculation elements are arranged so as to have the same array as the neuron model 30 in the neural network 26 shown in FIG. , Each of the product-sum calculation elements is configured to perform the same calculation as that of the neuron model 30 shown in FIG. At this time, the operation constants (that is, the weighting factors w1, w2, ..., Wi and the bias value α) of each product-sum operation element are the same as those described above in the neuron model arranged at the same position as the product-sum operation element. Set it to the same value as the operation constant determined by learning.

By doing so, the product-sum calculation circuit 46 has the same configuration and function as the neural network 26 after learning.

Therefore, as shown in FIG.
When the G signal actually obtained by the above is subjected to interval integration by the interval integration circuit 43, serial-parallel conversion is performed by the serial-parallel conversion circuit 45, and input to the product-sum operation circuit 46, at the time of a vehicle collision, Collision mode (ie collision mode such as frontal collision, oblique collision, offset collision, underride collision and pole collision, collision target such as A and B, vehicle speed at collision such as constant speed, medium speed and high speed) According to each combination), a signal close to an ideal output signal can be output from each output terminal of the product-sum calculation circuit 46.

Specifically, in the present embodiment, as in the case of the neural network 26 shown in FIG. 3, among the 11 output terminals of the product-sum calculation circuit 46, the uppermost output terminal is
A signal α indicating that the airbag device is not activated (squib ignition off) is output. The second and subsequent signals are, in order from the top, a signal β1 indicating that the collision mode is a normal collision and the collision target is A, and the collision mode is a pole collision and the collision target is A signal β9 indicating that the collision is A, and a signal β10 indicating that the collision mode is a pole collision and the collision target is B are output. Among them, the signal α indicates that the airbag device 50 should be activated when the signal value is “1”, and otherwise (that is, the airbag device 50 is activated when the signal value is “0”). It is not necessary to start). In addition, the signals β1 to β10
As for the signal β1, for example, when the signal value is “1”, it indicates that the collision mode is a head-on collision and the collision target is A, and when the signal value is “0”, other than that. Shows the case.

In this way, the plurality of signals output from the product-sum calculation circuit 46 are transmitted to the OR circuit 66 as described above.
And the AND circuit 64. Specifically, as shown in FIG. 1, of the 11 output terminals of the product-sum calculation circuit 46, it is not necessary to activate the airbag device output from the uppermost output terminal (squib ignition). Only the signal α indicating OFF) is inversely input to one input end of the AND circuit 64, and the ten signals β1 to β10 indicating the collision mode of the vehicle, which are respectively output from the remaining output ends, are output from the OR circuit 66. It is input to each input terminal.

Then, in the OR circuit 66, the input 10
The logical sum of these signals is obtained, and the logical sum signal is directly input to the other input terminal of the AND circuit 64. The AND circuit 64 obtains a logical product of the two input signals, and the logical product signal is sent to the drive circuit 52 to the airbag device 5
It is input as a 0 control signal. The start control signal is
When the signal value is "1", the airbag device 50 is activated, but when the signal value is "0", the airbag device 50 is not activated.

Therefore, for example, when a vehicle collision occurs, the mode is a head-on collision, and the collision object is close to A, the vehicle speed at the time of the collision is high and the airbag device is to be activated, the sum of products is calculated. Signals α, β1 to β output from the arithmetic circuit 46
Since the signal value of 10 is (0, 1, 0, 0, 0, ..., 0), the signal value of the logical sum signal from the OR circuit 66 is “1”, and the start signal output from the AND circuit 64 is output. The signal value of the control signal (logical product signal) also becomes "1", and the airbag device 50 is activated. Further, even when a vehicle collision occurs and the mode is a pole collision and the collision target is close to B, if the vehicle speed at the time of the collision is low and it is not necessary to start the airbag device, The signal values of the signals α, β1 to β10 output from the arithmetic circuit 46 are (1, 0,
0, 0, 0, ..., 1), the signal value of the logical sum signal from the OR circuit 66 becomes “1”, but the AND circuit 64
The signal value of the activation control signal (logical product signal) output from the device remains "0" and the airbag device 50 is not activated.

It should be noted that, in a special case that is not expected during learning, that is, the signal value of the signal α output from the product-sum calculation circuit 46 is a value "0" indicating that the airbag device should be activated. However, the signals β1 to β indicating the collision mode of the vehicle
When the signal values of 10 are all "0", the OR circuit 6
Since the signal value of the logical sum signal from 6 becomes "0", the activation control signal (logical product signal) output from the AND circuit 64
The signal value of “0” remains “0” and the airbag device 50
Does not start.

Next, the drive circuit 52 is connected to the start control device 40.
While the activation control signal output from is input and the signal value is “0”, the airbag device 50 is not activated, but when the signal value becomes “1”, the airbag device 50 is activated. The squib 51 in the airbag device 50 is energized.

The airbag device 50 includes a squib 51, a gas generating agent (not shown), a bag (not shown), and the like. Therefore, when the squib 51 is energized by the drive circuit 52, the squib 51 ignites the gas generating agent, whereby gas is generated from the gas generating agent. The gas inflates the bag and protects the occupant from collision with interior parts in the vehicle 32.

As described above, according to the present embodiment, whether the product-sum calculation circuit 46 should activate the signal indicating the collision mode of the vehicle or the airbag device 50 from the section integration signal which is time series data. Since it is possible to obtain a signal indicating whether or not the vehicle is in a collision state, it is possible to reliably determine whether or not the airbag device 50 should be activated.

In the present embodiment, the output end of the product-sum calculation circuit 46 can output not only a signal indicating whether or not the airbag device 50 should be activated, but also a signal indicating a collision mode of the vehicle. By using the signal indicating the collision mode, it is possible to easily check in advance whether the product-sum calculation circuit 46 works effectively. Further, when a vehicle collision occurs, the signal indicating the above-described collision mode is stored in a memory or the like, so that it is possible to easily verify in what collision mode the vehicle collision occurred after the vehicle collision.

FIG. 5 is a block diagram showing the construction of a start-up control device for an air bag device as a second embodiment of the present invention. The start control device 60 of the present embodiment is different from the first embodiment described above in that the AND circuit 64 and the OR circuit 66 are deleted, and the top of the 11 output terminals of the product-sum calculation circuit 46 is removed. The point is that only the signal α indicating that the airbag device output from the output end is not activated (squib ignition off) is directly input to the drive circuit 52 as the activation control signal of the airbag device 50. That is, since the signal α is a signal indicating whether or not the airbag device 50 should be activated, there is no problem even if it is used as it is as the activation control signal of the airbag device 50. However, the activation control signal is the reverse of that in the first embodiment. When the signal value is "0", the airbag device 50 is activated, but when the signal value is "1", the airbag device 50 is activated. Does not start.

In this embodiment, in a special case which is not assumed at the time of learning, that is, the signal value of the signal α output from the product-sum calculation circuit 46 described above should activate the airbag device. Even if the signal value indicating the effect is “0” and the signal values of the signals β1 to β10 indicating the collision mode of the vehicle are all “0”, the signal value of the start control signal is “1”. The bag device 50 is activated.

According to this embodiment, since the AND circuit 64 and the OR circuit 66 are eliminated, the number of parts can be reduced and the structure can be simplified.

By the way, in the two embodiments described above, only the G signal that will be obtained at the time of a vehicle collision is prepared as the G signal for learning, and when the vehicle is traveling on a rough road (this In this case, the G signal that would be obtained before the airbag device is activated is excluded.

The reason for excluding the G signal during traveling on a rough road is as follows. That is, the peak waveform of the G signal during traveling on a rough road is very similar to the peak waveform of the G signal at the initial stage of a collision at the time of a high-speed head-on collision, and it is difficult to determine. This is because when the signal is input as a signal and the neural network 26 is made to learn, the error between the output signal and the ideal output signal may not reach convergence to zero. When the G signal during traveling on a rough road is input as the G signal for learning, the neural network 2
From each output terminal of 6, from the top (1, 0, 0, 0, 0,
Since it is ideal that a signal of (1, 0) is output, a signal of (1, 0, 0, 0, 0, ..., 0) is prepared as an ideal output signal.

However, in this way, when the neural network 26 is trained by excluding the G signal during traveling on a rough road, each neuron model 3 finally obtained.
The arithmetic constant of 0 naturally does not include information about running on a rough road. Therefore, if these operation constants are used as the operation constants of the product-sum operation elements in the product-sum operation circuit 46 shown in FIG. 1, the operation constants of these product-sum operation elements also include information about running on a rough road. Therefore, in the start control device, it may be difficult to reliably determine whether the vehicle is traveling on a rough road or when a vehicle collides.

Therefore, an embodiment in which a rough road determination circuit 62 is provided separately from the product-sum calculation circuit 46 in order to reliably discriminate between running on a rough road and a vehicle collision will be described below.

FIG. 6 is a block diagram showing the arrangement of a start-up control device as a third embodiment of the present invention. In the startup control device 70 shown in FIG. 6, the same components as those in FIG. 1 are designated by the same reference numerals. Therefore, description of those components will be omitted.

The rough road determination circuit 62, which is a feature of this embodiment.
Inputs the G signal output from the A / D converter 42,
Based on the G signal, it is determined whether the vehicle is in a collision or traveling on a rough road, and the determination result is output as a determination signal. That is, when it is determined that the vehicle is in a collision, a signal having a value of "1" is output as the determination signal, and when it is determined that the vehicle is not in a collision (including during traveling on a bad road), the value is determined as the determination signal. The signal of "0" is output.

The AND circuit 64 is the product-sum operation circuit 4
A signal indicating that the airbag device output from the uppermost output end of 6 is not activated (squib ignition off) is inverted and input to one input end, and a logical sum signal from the OR circuit 66 is input. And the determination signal from the rough road determination circuit 62 are input to the other two input terminals respectively, the logical product of these is calculated, and the calculation result is output as a start control signal. As in the case of FIG. 1, the activation control signal output at this time activates the airbag device 50 when the signal value is “1”, but when the signal value is “0”, the airbag device 50 is activated. 5
Do not start 0.

Therefore, for example, if the value of the signal α from the product-sum calculation circuit 46 becomes "0" during running on a rough road, it indicates that the airbag device 50 should be started, and the logic from the OR circuit 66 is displayed. Even if the value of the sum signal is "1", the value of the determination signal from the bad road determination circuit 62 remains "0", indicating that the vehicle is not in a vehicle collision (that is, while traveling on a bad road). Therefore, the value of the output signal from the AND circuit 64 remains "0", and the airbag device 50 is not activated. Therefore, there is little possibility that the airbag device 50 will be activated during traveling on a rough road.

The rough road judging circuit 62 can be formed by, for example, an integrating circuit and a comparing circuit. In general, the integrated value of the G signal is relatively large during a vehicle collision, while it is relatively small during traveling on a rough road. Therefore, the input G signal is integrated by the integrating circuit, and the obtained integrated value is compared with a predetermined threshold value by the comparing circuit. When the integrated value is larger than the threshold value, it is determined that the vehicle is in a collision, and the determination signal of the value "1" is output. When the integrated value is smaller than the threshold value, other than the vehicle collision (even while traveling on a rough road. It is determined that the value is "included", and a determination signal of the value "0" is output.

Further, the rough road determination circuit 62 uses, for example, an LP
It is also possible to configure with an F circuit and a comparison circuit.
In general, a G signal does not include a high frequency component when a vehicle collides, whereas a G signal includes a large amount of a high frequency component during traveling on a rough road. Therefore, from the input G signal to the LPF
The circuit removes high frequency components, and the value of the G signal is compared with a predetermined threshold value by a comparison circuit. When the value of the G signal is larger than the threshold value, it is determined that the vehicle is in collision, and the determination signal of the value “1” is output. (Including driving on the road)
Then, the determination signal of the value "0" is output.

As described above, according to the present embodiment, since the bad road determination circuit 62 is provided, it is possible to reliably determine whether the vehicle is traveling on a bad road or when a vehicle collides. The possibility of activating the airbag device 50 is low.

FIG. 7 is a block diagram showing the construction of a start-up control device for an air bag device according to a fourth embodiment of the present invention. In the activation control device 80 of the present embodiment, the sum-of-products calculation circuit 46 indicates that it is not necessary to activate the airbag device from the uppermost output end of the 11 output ends (squib ignition off). The output signal is the signal α, but the signals β1 to β10 indicating the collision mode of the vehicle are output from the second and subsequent output ends.
Is output. Therefore, in the present embodiment, a plurality of airbag devices are prepared, and as described above, the signals β1 to β10 indicating the vehicle collision mode output from the second and subsequent output ends of the product-sum calculation circuit 46 are used, The activation of each airbag device is controlled according to the type of collision of the vehicle. Therefore, in this embodiment, as shown in FIG.
In addition to the configuration shown in FIG.
And airbag devices 54 and 56 are added respectively.

In FIG. 7, the air bag devices 50, 5
Reference numerals 4 and 56 denote airbag devices mounted on the vehicle 32. For example, if the airbag device 50 is a driver airbag device, the airbag device 54 is replaced by a passenger airbag device and an airbag device. 56 may be a rear seat airbag device. Further, for example, when the airbag device 50 is a driver's seat front-collision airbag device, the airbag device 54 is a driver-seat side-collision airbag device, and the airbag device 56 is a passenger's-seat front-collision airbag device. can do.

The drive circuits 52, 55, 57 are provided corresponding to the airbag devices 50, 54, 56, respectively, and as will be described later, the activation control signal output from the selection circuit 68 is input and the signal values thereof are input. While the corresponding airbag device is not activated while is 0, the squib in the airbag device is energized to activate the corresponding airbag device when the signal value becomes "1".

The selection circuit 68 inputs the signals α, β1 to β10 output from the respective output terminals of the product-sum calculation circuit 46, and determines whether to activate any of the airbag devices based on the signal α. In addition to the above determination, when the collision mode of the vehicle is determined by the signals β1 to β10 and the airbag apparatus is to be activated, a plurality of airbag apparatuses 50, 54, 54, 54, Of the 56, which airbag device is to be activated is selected, and the drive circuit corresponding to the selected airbag device has a signal value “1” indicating that the airbag device should be activated as an activation control signal. The signal of is output.

As described above, according to the present embodiment, the signals β1 to β10 indicating the collision mode of the vehicle output from the second and subsequent output terminals of the product-sum calculation circuit 46 are used to determine the collision mode of the vehicle. Accordingly, an appropriate airbag device among the plurality of airbag devices can be properly activated.

The present invention is not limited to the above-described examples and embodiments, and can be implemented in various modes without departing from the scope of the invention.

In the above-described embodiment, the uppermost output end of the 11 output ends of the neural network 26 or the product-sum calculation circuit 46 is not sufficient to activate the airbag device (squib ignition). OFF) is output, and the second and subsequent ones are five modes of normal collision, oblique collision, offset collision, underride collision, and pole collision, and A and B.
The signals indicating the collision mode of the vehicle are output corresponding to the respective combinations of the two collision objects. However, the present invention is not limited to this, and for example, a signal indicating the vehicle speed (constant speed, medium speed, high speed, etc.) at the time of collision may be additionally added and output. Further, it is also possible to output only the mode of collision, only the collision object, or only the vehicle speed at the time of collision without combining them.

In the above-described embodiments, the product-sum calculation circuit 46 used in the activation control device has been described on the assumption that a general product-sum calculation circuit is used. However, the neural network also has a learning function. A circuit,
Since it is a kind of product-sum operation circuit, it can be used as the product-sum operation circuit 46. In this case, the learned neural network 2 used in the product-sum calculation circuit learning device 20
Although 6 may be used as it is, another neural network may be prepared, and an operation constant obtained by learning may be set in the neural network and used.

In the above embodiment, the G signal is subjected to the interval integration as the vehicle impact signal corresponding to the magnitude of the impact applied to the vehicle, and the interval integration signal is input to the serial / parallel conversion circuit. In the present invention, the G signal may be directly input to the serial / parallel conversion circuit as the vehicle impact signal without being limited to the section integration signal.
A signal obtained by performing a predetermined calculation (for example, second-order integration or differentiation) other than the interval integration on the signal may be input to the serial / parallel conversion circuit as a vehicle impact signal.

In the above-mentioned embodiment, the airbag device is used as the occupant protection device. However, other than the airbag device, for example, an auto pretensioner or an inflatable curtain may be used. The auto pretensioner operates, for example, to wind up a seat belt with gas generated by an inflator and further enhance the function of the seat belt. In addition, the inflatable curtain is attached above the inside of the door in the driver's seat and the like, and descends like a curtain by the gas generated by the inflator to protect the occupant's head and the like.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a start-up control device for an airbag device according to a first embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a product-sum calculation circuit learning device for determining an arithmetic constant of a product-sum calculation element in the product-sum calculation circuit of FIG. 1 by learning.

3 is a configuration diagram showing an internal configuration of a neural network 26 shown in FIG.

4 is a neuron model 30 shown in FIG. 3 and a sigmoid function fs used in the neuron model 30;
It is an explanatory view for explaining (X).

FIG. 5 is a block diagram showing a configuration of a start-up control device for an airbag device as a second embodiment of the present invention.

FIG. 6 is a block diagram showing a configuration of a start-up control device for an airbag device according to a third embodiment of the present invention.

FIG. 7 is a block diagram showing a configuration of a start-up control device for an airbag device according to a fourth embodiment of the present invention.

[Explanation of symbols]

 20 ... Product-sum operation circuit learning device 22 ... A / D converter 23 ... Section integration circuit 25 ... Serial / parallel conversion circuit 26 ... Neural network 28 ... Adjustment circuit 30 ... Neuron model 32 ... Vehicle 40 ... Startup control device 41 ... G Sensor 42 ... A / D converter 43 ... Section integration circuit 45 ... Serial / parallel conversion circuit 46 ... Sum-of-products arithmetic circuit 50 ... Airbag device 51 ... Squib 52 ... Drive circuit 54 ... Airbag device 55 ... Drive circuit 56 ... Air Bag device 57 ... Driving circuit 60 ... Startup control device 62 ... Bad road determination circuit 64 ... AND circuit 66 ... OR circuit 68 ... Selection circuit 70 ... Startup control device 80 ... Startup control device

Front Page Continuation (72) Inventor Hideki Hashimoto 7-22-1, Roppongi, Minato-ku, Tokyo Inside the Institute of Industrial Science, University of Tokyo

Claims (11)

[Claims] 1. An occupant protection device is started by inputting a signal obtained from a vehicle impact signal corresponding to the magnitude of an impact applied to a vehicle and performing a sum of products operation using a plurality of arithmetic constants on the signal. A product-sum operation circuit learning method for determining the plurality of operation constants by learning, for a product-sum operation circuit for a start control device that outputs a signal for obtaining a start control signal for controlling A step of preparing a learning product-sum operation circuit having a plurality of product-sum operation elements, having a plurality of input terminals and a plurality of output terminals, and capable of adjusting operation constants of the plurality of product-sum operation elements; , (B) converting the learning vehicle impact signal, which is a time-series signal prepared in advance, into series-parallel conversion, and (c) converting the learning vehicle impact signal subjected to serial-parallel conversion into the learning product-sum operation circuit. Parallel input to each input terminal And a step of performing a product-sum operation using the operation constants by the plurality of product-sum operation elements in the learning product-sum operation circuit, and (d) each output terminal of the learning product-sum operation circuit as an operation result. The learning sum-of-products arithmetic circuit so that each error between the plurality of output signals output from the plurality of output signals and the plurality of ideal output signals prepared in advance corresponding to the learning vehicle impact signal has a predetermined value. A step of adjusting the operation constants of the plurality of product-sum operation elements in the step of: A method for learning a product-sum operation circuit, characterized in that it is determined as the plurality of operation constants. 2. The product-sum calculation circuit learning method according to claim 1, wherein the step (a) includes a step of preparing a neural network as the learning product-sum calculation circuit. Method. 3. The product-sum operation circuit learning method according to claim 1, wherein said step (b) is obtained as a vehicle impact signal for learning, each of which is obtained at the time of a vehicle collision and has different collision modes. And a step (d) of preparing the plurality of ideal output signals which are different for each of the plurality of vehicle shock signals. 4. The product-sum operation circuit learning method according to claim 1, wherein the step (b) is a vehicle other than a vehicle impact signal obtained during traveling on a rough road of the vehicle as the learning vehicle impact signal. A method of learning a product-sum operation circuit, comprising a step of preparing an impact signal. 5. An occupant protection device is activated by inputting a signal obtained from a vehicle impact signal corresponding to the magnitude of an impact applied to the vehicle and performing a sum of products operation using a plurality of arithmetic constants on the signal. A product-sum operation circuit learning device for determining the plurality of operation constants by learning, regarding a product-sum operation circuit for a start control device that outputs a signal for obtaining a start control signal for controlling Serial-parallel conversion means for converting the learning vehicle impact signal, which is a time-series signal, into serial-parallel conversion, a plurality of multiply-accumulate operation elements, and a plurality of input ends and a plurality of output ends. The learning vehicle impact signal is input in parallel from the input end, the product-sum calculation is performed using the calculation constants by the plurality of product-sum calculation elements, and the calculation constants of the plurality of product-sum calculation elements are adjusted. To do Capable learning product-sum operation circuit, an output signal output from each output terminal of the learning product-sum operation circuit as an operation result, and a plurality of ideal output signals prepared in advance corresponding to the learning vehicle impact signal And an arithmetic constant adjusting means for adjusting arithmetic constants of the plurality of product-sum operation elements in the learning product-sum operation circuit so that each error becomes a predetermined value. A product-sum operation circuit learning device, wherein the operation constants of the plurality of product-sum operation elements are determined as the plurality of operation constants in the product-sum operation circuit for the startup control device. 6. An activation control device for an occupant protection device, which is a time-series signal, which obtains an activation control signal for controlling the activation of the occupant protection device based on a vehicle impact signal corresponding to the magnitude of the impact applied to the vehicle. Serial-parallel conversion means for converting the vehicle impact signal in series-parallel, having a plurality of input ends and a plurality of output ends, the vehicle impact signal subjected to serial-parallel conversion is input in parallel from the input end,
A sum-of-products operation circuit that performs a sum-of-products operation using a plurality of operation constants and outputs signals obtained as the operation results from the respective output terminals; While obtaining the activation control signal, as the plurality of operation constants in the product-sum operation circuit,
An activation control device for an occupant protection device, which uses an operation constant determined by the product-sum operation circuit learning method according to claim 1, 2, or 3. 7. The activation control device for an occupant protection device according to claim 6, wherein one of the plurality of signals output from each output terminal of the product-sum calculation circuit is the occupant protection device. An activation control device for an occupant protection device, which is a signal for determining whether or not to activate the device. 8. The activation control device for an occupant protection system according to claim 6, wherein the plurality of signals output from the output terminals of the product-sum calculation circuit are output when the vehicle collides. An activation control device for an occupant protection system including a signal capable of identifying a collision mode. 9. The activation control device for an occupant protection device according to claim 6, wherein the vehicle impact signal is a signal obtained by performing interval integration of acceleration applied to the vehicle. Device startup control device. 10. The activation control device for an occupant protection system according to claim 6, wherein a determination signal for determining whether the vehicle is in a collision or is traveling on a rough road based on the vehicle impact signal. Determining means to be derived, and the signal output from the product-sum calculation circuit indicates that the occupant protection device should be activated, and the determination signal derived by the determining means to indicate a vehicle collision The activation control signal, the activation control signal output means outputs a signal for activating the occupant protection device as the activation control signal. 11. A startup control device for an occupant protection device, which is a time-series signal, which obtains a startup control signal for controlling the startup of the occupant protection device based on a vehicle impact signal corresponding to the magnitude of the impact applied to the vehicle. Serial-parallel conversion means for converting the vehicle impact signal in series-parallel, having a plurality of input ends and a plurality of output ends, the vehicle impact signal subjected to serial-parallel conversion is input in parallel from the input end,
A sum-of-products calculation circuit that performs a sum-of-products calculation using a plurality of calculation constants and outputs signals obtained as the calculation results from the respective output terminals, and a vehicle collision or a bad road based on the vehicle-impact signal. Discriminating means for deriving a discriminating signal for discriminating whether or not the vehicle is running, and the signal output from the product-sum calculation circuit indicates that the occupant protection device should be activated, and the discriminating means A start-up control signal output means for outputting a signal for starting up the occupant protection device as the start-up control signal when the derived discrimination signal indicates a vehicle collision; As the plurality of arithmetic constants in
An activation control device for an occupant protection device, which uses an operation constant determined by the product-sum operation circuit learning method according to claim 4.