Method for machine failure prediction using memory depth values
Embodiments of the invention provide a method and system for machine failure prediction. The method comprises: identifying a plurality of basic memory depth values based on a composite sequence of machine failure history; ascertaining weight values for at least one of the identified basic memory depth values according to a pre-stored table which includes a plurality of mappings wherein each mapping relates a basic memory depth value to one set of weight values; and predicting a future failure using a Back Propagation Through Time (BPTT) trained Recurrent Neural Network (RNN) based on the ascertained weight values, wherein weight values related to a first basic memory depth value in the pre-stored table is ascertained based on a second set of weight values related to a second basic memory depth value which is less than the first basic memory depth value by a predetermined value.
1. A method of machine failure prediction, the method comprising:
determining, by at least one processor in at least one computer system, a first basic memory depth value and a second basic memory depth value based on a composite sequence of a machine failure history;
inputting, by the at least one processor, the first basic memory depth value, a first predetermined error threshold, and a first set of weight values between a first predetermined range into a recurrent neural network;
running, by the at least one processor, the recurrent neural network based on the first basic memory depth value, the first predetermined error threshold, the first set of weight values, and a back propagation of errors training method until a first final set of weight values converge to predict a first elementary sequence corresponding to the first basic memory depth value;
inputting, by the at least one processor, the second basic memory depth value, a second predetermined error threshold, and a second set of weight values between a second predetermined range into the recurrent neural network;
running, by the at least one processor, the recurrent neural network based on the second basic memory depth value, the second predetermined error threshold, the second set of weight values, and the back propagation of errors training method until a second final set of weight values converge to predict a second elementary sequence corresponding to the second basic memory depth value;
determining, by the at least one processor, a set of initial composite weight values based on a weighted average of the first final set of weight values and the second final set of weight values;
inputting, by the at least one processor, the set of initial composite weight values and a third predetermined error threshold into the recurrent neural network; and
running, by the at least one processor, the recurrent neural network based on the set of initial composite weight values, the third predetermined error threshold, and the back propagation of errors training method until a third final set of weight values converge to predict a future failure.
2. The method according to claim 1 , further comprising:
determining, by the first processor, a number of state units to be used by the recurrent neural network based on a maximum value of the determined first and second basic memory depth values.
3. The method according to claim 2 , wherein the number of state units is further determined based on the following rules:
if the maximum value of the determined first and second basic memory depth values is not greater than 350, the number of state units is 60 units;
if the maximum value of the determined first and second basic memory depth values is more than 350 units, but not greater than 1000 units, the number of state units is 120 units.
4. The method according to claim 1 , wherein the first set of weight values related to the first basic memory depth value is based on the second set of weight values related to the second basic memory depth value, and wherein the second basic memory depth value is less than the first basic memory depth value by a predetermined value.
5. The method according to claim 4 , wherein the predetermined value is an integer not greater than 5.
6. A system of machine failure prediction, comprising: at least one processor and at least one memory communicably coupled thereto,
wherein the at least one memory is configured to store data to be executed by the at least one processor,
wherein the at least one processor is configured to:
determine a first basic memory depth value and a second basic memory depth value based on a composite sequence of a machine failure history;
input, by the at least one processor, the first basic memory depth value, a first predetermined error threshold, and a first set of weight values between a first predetermined range into a recurrent neural network;
run, by the at least one processor, the recurrent neural network based on the first basic memory depth value, the first predetermined error threshold, the first set of weight values, and a back propagation of errors training method until a first final set of weight values converge to predict a first elementary sequence corresponding to the first basic memory depth value;
input, by the at least one processor, the second basic memory depth value, a second predetermined error threshold, and a second set of weight values between a second predetermined range into the recurrent neural network;
run, by the at least one processor, the recurrent neural network based on the second basic memory depth value, the second predetermined error threshold, the second set of weight values, and the back propagation of errors training method until a second final set of weight values converge to predict a second elementary sequence corresponding to the second basic memory depth value;
determine, by the at least one processor, a set of initial composite weight values based on a weighted average of the first final set of weight values and the second final set of weight values;
input, by the at least one processor, the set of initial composite weight values and a third predetermined error threshold into the recurrent neural network; and
run the recurrent neural network based on the set of initial composite weight values, the third predetermined error threshold, and the back propagation of errors training method to predict a future failure.
7. The system according to claim 6 , wherein the at least one processor is further configured to determine a number of state units to be used by the recurrent neural network based on a maximum value of the determined first and second basic memory depth values.
8. The system according to claim 7 , wherein the at least one processor is further configured to determine the number of state units based on the following rules:
if the maximum value of the determined first and second basic memory depth values is not greater than 350, the number of state units is 60 units;
if the maximum value of the determined first and second basic memory depth values is more than 350 units, but not greater than 1000 units, the number of state units is 120 units.
9. The system according to claim 6 , wherein the first set of weight values related to the first basic memory depth value is based on the second set of weight values related to the second basic memory depth value, and wherein the second basic memory depth value is less than the first basic memory depth value by a predetermined value.
10. The system according to claim 9 , wherein the predetermined value is an integer not greater than 5.
11. A non-transitory computer readable medium comprising computer program code to perform machine failure prediction, wherein the computer program code, when executed, is configured to cause the at least one processor in the at least one computer system to perform the method according to claim 1 .