System and method for improving engineer-to-order configuration

1 . A method for determining a set of output configuration values characterizing a specific configuration of a complex product, the method comprising:

a) receiving a set of input configuration parameters that is orderless;

b) providing at least a part of the input configuration parameters as an input to a solver;

c) using the solver to calculate at least one output value from the provided input configuration parameters;

d) determining the set of output configuration values at least from the at least one output value calculated by the solver; and

e) using the solver for solving a first order logic function encoding an algorithm of a trained deep neural network, the algorithm of the deep neural network having been trained for modeling a function of an external configuration tool required for determining the specific configuration of the complex product;

f) the first order logic function, which encodes the algorithm of the trained deep neural network, enabling the solver to solve the first order logic function with the at least a part of the set of input configuration parameters being orderless;

g) using the system to receive as an input from a user the output configuration values required to be reached or satisfied by the complex product once the specific configuration of the complex product is achieved, and using the system to automatically determine constraints or requirements on the input configuration parameters from the output configuration values.

2 . The method according to claim 1 , which further comprises using the at least one output value as an input to another solver configured for solving another first order logic function encoding an algorithm of another deep neural network, the other deep neural network having been trained for approximating another function of the external configuration tool or a function of another external configuration tool.

3 . The method according to claim 1 , which further comprises automatically transmitting the output configuration values to a manufacturing system configured for controlling manufacturing processes according to the output configuration values.

4 . The method according to claim 1 , which further comprises providing the deep neural network with a rectified linear unit as an activation function.

5 . A method for training a deep neural network to approximate a function of an external configuration tool, the method comprising:

receiving sets of input parameters, each set of input parameters including all parameters needed by a first order logic function of the external configuration tool for calculating an associated set of one or several external configuration tool output values involved in a determination of a specific product configuration of a complex product;

receiving for each set of input parameters the associated set of one or several external configuration tool output values, the one or several external configuration tool output values having been obtained by running the function of the external configuration tool when using it as an input of a latter of the set of input parameters to which the one or several external configuration tool output values are associated;

training an algorithm of the deep neural network approximating the external configuration tool, the training being based on pair of sets including each one of the sets of input parameters and its associated set of output values; and

storing the trained deep neural network;

the trained deep neural network having an algorithm encoded by a first order logic function enabling the first order logic function to be solved by a solver provided with, as input, at least a part of a set of input configuration being orderless;

during data collection implemented by the training, using the external configuration tool to calculate for each set of the input parameters the associated set of the external configuration tool output values being automatically collected by the training and stored in a memory.

6 . A system for determining a set of output configuration values characterizing a specific configuration of a complex product, the system comprising:

a first interface for acquiring a set of input configuration parameters that is orderless;

a computation unit including a solver provided with input being at least a part of the set of input configuration parameters that is orderless; and

a second interface for providing output configuration values generated by said computation unit;

said solver configured for solving a first order logic function encoding an algorithm of a trained deep neural network, the algorithm of the deep neural network having been trained for modeling a function of an external configuration tool required for determining the specific configuration of the complex product;

the first order logic function, which encodes the algorithm of the trained deep neural network, enabling the solver to solve the first order logic function with the at least a part of the set of input configuration parameters being orderless;

the system configured to receive as an input from a user the output configuration values required to be reached or satisfied by the complex product once the specific configuration of the complex product is achieved, and the system configured to automatically determine constraints or requirements on the input configuration parameters from the output configuration values.

7 . The system according to claim 6 , which further comprises a manufacturing system including manufacturing machines configured for automatically controlling manufacturing processes according to the output configuration values to thereby obtain the complex product with the specific configuration characterized by the output configuration values, wherein said manufacturing machines include at least one manufacturing machine having a manufacturing tool configured for motion controlled by the output configuration values.

8 . A training system configured for training a deep neural network to approximate a function of an external configuration tool, the training system comprising:

a first training interface configured for acquiring sets of input parameters;

a second training interface configured for receiving, for each set of input parameters, an associated set of external configuration tool output values generated by executing the function of the external configuration tool;

a training computation unit configured for training the deep neural network based on each pair of sets including a set of input parameters and its associated set of external configuration tool output values, said training computation unit configured for controlling the external configuration tool to cause the external configuration tool to provide for each set of input parameters the set of external configuration tool output values, the set of external configuration tool output values being collected by said training computation unit through said second training interface in order to train the deep neural network; and

a memory configured for storing the trained deep neural network;

the trained deep neural network having an algorithm encoded by a first order logic function enabling the first order logic function to be solved by a solver provided with, as input, at least a part of a set of input configuration parameters being orderless;

during data collection implemented by the training system, the external configuration tool calculating for each set of the input parameters the associated set of the external configuration tool output values being automatically collected by the training system and stored in the memory.