Fixed-point virtual sensor control system and method

• US 20090112334A1
• Filed: 10/31/2007
• Published: 04/30/2009
• Est. Priority Date: 10/31/2007
• Status: Active Grant

• Current Assignee: Caterpillar Incorporated
• Original Assignee: Caterpillar Incorporated
• Inventors: Mason, James, Grichnik, Anthony J., Felty, Tim
• Granted Patent: US 7,593,804 B2
• Time in Patent Office: Days
• US Class Current: 700/29
• CPC Class Codes: F02D 41/1405 Neural network control
• CPC Class Codes: F02D 41/1406 with use of a optimisation method, e.g. iteration
• CPC Class Codes: F02D 41/26 using computer, e.g. microprocessor
• CPC Class Codes: G05B 23/0221 Preprocessing measurements, e.g. data collection rate adjustment; Standardization of measurements; Time series or signal analysis, e.g. frequency analysis or wavelets; Trustworthiness of measurements; Indexes therefor; Measurements using easily measured parameters to estimate parameters difficult to measure; Virtual sensor creation; De-noising; Sensor fusion; Unconventional preprocessing inherently present in specific fault detection methods like PCA-based methods

Abstract

One aspect of the present disclosure includes a method for a control system of a machine. The method may include establishing a virtual sensor model indicative of interrelationships between at least one sensing parameter and a plurality of measured parameters related to the machine. The method may also include obtaining data and function information representing the virtual sensor model and converting the data information into fixed-point representation. Further, the method may include converting the function information into fixed-point representation and loading the converted fixed-point representation of data information and function information in the control system such that the control system uses the virtual sensor model in fixed-point arithmetic operation.

Claims

1. A method for a control system of a machine, comprising:

establishing a virtual sensor model indicative of interrelationships between at least one sensing parameter and a plurality of measured parameters related to the machine;

obtaining data and function information representing the virtual sensor model;

converting the data information into fixed-point representation;

converting the function information into fixed-point representation; and

loading the converted fixed-point representation of data information and function information in the control system such that the control system uses the virtual sensor model in fixed-point arithmetic operation to control the machine.

2. The method according to claim 1, wherein:

the virtual sensor model is a neural network virtual sensor model; and

the virtual sensor model includes a plurality of neural network layers, each of which includes one or more neural node.

3. The method according to claim 2, wherein:

the neural node includes at least one weight for a connection between the neural node and a different neural node, and an activation function; and

the data information includes at least the weight and a gain and offset of the activation function.

4. The method according to claim 3, wherein converting the data information includes:

determining a fixed-point data type for each number included in the data information based on a magnitude and precision of each number; and

converting each number into a fixed-point number according the respective fixed-point data type.

5. The method according to claim 2, wherein:

the neural node includes an activation function; and

the function information includes a type, input and output relationship, and input and output ranges of the activation function.

6. The method according to claim 5, wherein converting the function information include:

obtaining the input range of the activation function;

generating a fixed-point function map representing the input and output relationship of the activation function;

determining the output range of the activation function; and

determining at least one fixed-point data type corresponding to the output range.

7. The method according to claim 6, wherein the function type of the activation function is one of an identity function, an exponential function, a hyperbolic tangent function, and a sigmoidal function.

8. The method according to claim 6, wherein:

the function type of the activation function is an exponential function; and

the at least one fixed-point data type includes a plurality of fixed-point data types each corresponding to a respective part of the output range determined by the input range and the fixed-point function map.

9. The method according to claim 2, further including:

obtaining values of the plurality of measured parameters;

providing the obtained values to the virtual sensor model; and

obtaining fixed-point value of the at least one measuring parameter from the virtual sensor model through fixed-point arithmetic operation.

10. The method according to claim 9, wherein:

the at least one measuring parameter includes one of a NO_x emission level, a soot emission level, and an HC emission level; and

the plurality measured parameters include at least engine speed, fuel rate, injection timing, intake manifold temperature, intake manifold pressure, inlet valve actuation end of current, and injection pressure.

11. A computer system for converting a virtual sensor model to fixed-point representation used by a control system of a machine, comprising:

a database configured to store information relevant to the virtual sensor model; and

a processor configured to;

establish the virtual sensor model indicative of interrelationships between at least one sensing parameter and a plurality of measured parameters related to the machine;

obtain data and function information representing the virtual sensor model;

convert the data information into fixed-point representation;

convert the function information into fixed-point representation; and

load the converted fixed-point representation of data information and function information in the control system such that the control system uses the virtual sensor model in fixed-point arithmetic operation to control the machine.

12. The computer system according to claim 11, wherein:

the virtual sensor model is a neural network virtual sensor model; and

the virtual sensor model includes a plurality of neural network layers, each of which includes one or more neural node.

13. The computer system according to claim 12, wherein:

the neural node includes a weight for a connection between the neural node and a different neural node, and an activation function; and

the data information includes at least the weight and a gain and offset of the activation function.

14. The computer system according to claim 13, wherein, to convert the data information, the processor is configured to:

determine a fixed-point data type for each number included in the data information based on a magnitude and precision of each number; and

convert each number into a fixed-point number according the respective fixed-point data type.

15. The computer system according to claim 12, wherein:

the neural node includes an activation function; and

the function information includes a type, input and output relationship, and input and output ranges of the activation function.

16. The computer system according to claim 15, wherein, to convert the function information, the processor is configured to:

obtain the input range of the activation function;

generate a fixed-point function map representing the input and output relationship of the activation function;

determine the output range of the activation function; and

determine at least one fixed-point data type corresponding to the output range.

17. The computer system according to claim 16, wherein the function type of the activation function is one of an identity function, an exponential function, a hyperbolic tangent function, and a sigmoidal function.

18. The computer system according to claim 16, wherein:

the function type of the activation function is an exponential function; and

the at least one fixed-point data type includes a plurality of fixed-point data types each corresponding to a respective part of the output range determined by the input range and the fixed-point function map.