Surgical generator for ultrasonic and electrosurgical devices

A method of operating a surgical device, the method comprising:

• providing a drive signal to the surgical device to drive an end effector of the surgical device;
• receiving a frequency of the drive signal;
• and
• determining a temperature at the end effector considering the frequency of the drive signal, wherein determining the temperature at the end effector comprises:
• receiving a set of input variables selected from a portion of a plurality of input variables, the selected set of input variables including the frequency of the drive signal;
• weighting at least a portion of the input variables selected from the set of input variables to generate a weighted set of input variables;
• summing the weighted set of input variables to generate a first hidden layer node output;
• receiving a set of outputs from hidden layer nodes, wherein the set of outputs from the hidden layer nodes comprises the first hidden layer node output;
• weighting at least a portion of the set of outputs from the hidden layer nodes to generate a weighted set of outputs from the hidden layer nodes;
• summing the weighted set of outputs from the hidden layer nodes;
• and
• applying a transform function to the sum of the weighted set of outputs from the hidden layer nodes to generate a second layer node output.

A surgical system, comprising:

• a surgical device, wherein the surgical device comprises an end effector to treat tissue;
• and
• a surgical generator in communication with the surgical device, the surgical generator comprising at least one programmable logic device programmed to:
• provide a drive signal to drive the end effector;
• receive a frequency of the drive signal;
• and
• determine a temperature at the end effector considering the frequency of the drive signal and training a neural network model of end effector temperature, wherein training the neural network model comprises:
• providing to the model a set of input values including the frequency of the drive signal and changing or varying weight values w, bias values b, and transform functions f of the set of input values such that an output temperature of the neural network model approximates a measured dependency of the output temperature for known values of the set of input values;
• comparing the output temperature of the neural network model to a known output temperature associated with the set of input values;
• and
• modifying the weight values w, the bias values b, or the transform functions f of the set of input values until an error between the output temperature of the neural network model and a corresponding measured output temperature is below a predetermined error level.