JP2008009934A - Data processor, data processing method, remote diagnostic system for working machine and remote diagnostic method for woriking machine - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To more accurately and efficiently analyze the status of an object in simple configurations in a data processor, a data processing method, the remote diagnostic system of a working machine and the remote diagnostic method of the working machine. <P>SOLUTION: This data processor is provided with a parameter detection means 1 for detecting a plurality of parameters varying according to the status of an object; a parameter group classification means for classifying the plurality of parameters detected by the parameter detection means 1 into a first parameter group and a second parameter group; a neuron compression means 6 for compressing multi-dimensional actual data constituted of the parameters classified into the first parameter group into neuron by unsupervised learning of a neural network; and a basic statistic calculation means 7 for calculating the basic statistics of each of the parameters classified into the second parameter group. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention uses an unsupervised learning technique of a neural network to diagnose the state of various objects such as various machines, life forms such as animals and plants, microorganisms, and natural phenomena such as weather and celestial motion. In particular, the present invention relates to a work machine remote diagnosis system and a work machine remote diagnosis method suitable for use in diagnosing an operating state generated in a work machine such as a hydraulic excavator.

Conventionally, in order to analyze the structure of a huge number of data, a technique for compressing data into neurons using a method of unsupervised learning of a neural network, or restoring the original data structure from compressed neurons Technology has been developed.
For example, Patent Document 1 discloses a configuration in which a plurality of data sets including a plurality of parameter values that vary according to the operating state of a machine are detected by a sensor attached to a target machine, or a plurality of data sets are stored in a neuron. A configuration that compresses into neuron model parameters including coordinate information, average distance information, and weight information, and sends these neuron model parameters from the machine to the management center to calculate the moving average of the neurons and the density distribution (distribution density) of the data set The structure etc. which require and analyze a data set are indicated. With such a configuration, the machine diagnosis can be performed by more accurately reproducing the characteristics of the original data set from the compressed data.
JP 2006-11849 A

  However, when there are a large number of parameters that vary depending on the operating state of the machine, that is, when the number of data sets to be compressed into neurons is enormous, the time required for compression of these data sets Labor will also increase. For example, when making a diagnosis on a work machine such as a hydraulic excavator, not only parameters related to the engine state, such as engine speed, supercharging pressure to the engine, fuel consumption rate, exhaust temperature, etc., but also the power of the hydraulic circuit Various parameters such as shift pressure, hydraulic pump pressure, load pressure, hydraulic oil temperature, hydraulic pump case pressure, hydraulic pump drain pressure and other parameters related to the hydraulic circuit and hydraulic pump, as well as outside temperature, total operating time, work mode, etc. It is necessary to consider.

Therefore, if a data set composed of all these parameter values is to be compressed into neurons, complicated arithmetic processing must be performed for a long time. Also, similar efforts are required to restore the original data set characteristics from the compressed data.
The present invention has been made in view of such problems, and is a data processing apparatus, a data processing method, and an operation that can analyze the state of an object more accurately and efficiently with a simple configuration. An object of the present invention is to provide a machine remote diagnosis system and a work machine remote diagnosis method.

  In order to achieve the above object, a data processing apparatus according to a first aspect of the present invention includes a parameter detection means (data detection apparatus) that detects a plurality of parameters that vary in accordance with the state of an object, and the parameter detection means. A parameter group classifying means (calculation condition setting unit) that classifies the detected parameters into a first parameter group and a second parameter group, and a multi-dimensional actual parameter including parameters classified into the first parameter group. Neuron compression means (neuron compression unit) that compresses data into neurons by unsupervised learning of a neural network, and basic statistic calculation means (basic statistic calculation means that calculates basic statistics of each parameter classified into the second parameter group And a statistic calculation unit).

  In other words, instead of compressing all detected information into neurons, it classifies the detected information into information used for neuron compression and information used for simpler arithmetic processing, and only the former information is sent to neurons. And compress. The latter information calculates a basic statistic that is easier to calculate and can be processed in a short time, thereby reducing the amount of information. The basic statistic here refers to a statistic that represents the basic characteristics of quantity data. Here, representative values such as the number of data, average value of data, mode value, maximum value, and minimum value, This refers to the degree of dispersion such as standard deviation.

  In the case of the data processing apparatus related to the determination of the operating state of the work machine, the parameters detected by the parameter detecting means include, for example, total operating time information, operating position information, engine / vehicle error information, and accelerator dial information. , Information on lever operation amount, engine speed, engine oil pressure, boost pressure, fuel consumption, exhaust temperature, power shift pressure, hydraulic pump pressure, load factor (operation rate), engine oil temperature, engine coolant Temperature, hydraulic oil temperature, outside air temperature, hydraulic pump case temperature, hydraulic pump drain pressure, and the like.

According to a second aspect of the present invention, there is provided the data processing device according to the first aspect, wherein the parameter group classification means is configured to allow the plurality of parameters based on a predetermined condition relating to a time variation amount of each of the parameters. It is characterized by classifying.
For example, in the case of a data processing device related to the determination of the working state of a work machine, the parameter group classification means includes parameters [engine speed, engine oil pressure, boost pressure, fuel consumption, exhaust temperature, Power shift pressure, hydraulic pump pressure, load factor, etc.) are classified into the first parameter group, while parameters with relatively small time fluctuations (engine oil temperature, engine coolant temperature, hydraulic oil temperature, outside air) Temperature, hydraulic pump case temperature, hydraulic pump drain pressure, etc.) are classified into the second parameter group.

  According to a third aspect of the present invention, there is provided a data processing device according to the first or second aspect of the present invention, wherein, in the configuration according to the first or second aspect, the data processing device is set in advance among a plurality of multidimensional real data including parameters classified into the first parameter group. Real data extraction means (actual data extraction unit) for extracting real data including predetermined parameters satisfying the extraction conditions as effective real data, and neuron compression means is extracted by the real data extraction means. It is characterized by compressing effective real data into neurons by unsupervised learning of a neural network.

In other words, not all the information classified into the first parameter group is compressed into neurons, but only effective real data including predetermined parameters satisfying a preset extraction condition is compressed into neurons.
In the case of the data processing device related to the determination of the operating state of the work machine, as the extraction condition, for example, the actual data is detected at the time of operation at the three most frequently used accelerator dial positions, For example, the work machine is not in a warm-up operation, and the work machine is in a load operation (the engine of the work machine is not in an idling state). For example, whether the temperature information such as the engine oil temperature, the engine coolant temperature, the hydraulic oil temperature, the outside air temperature, the hydraulic pump case temperature is equal to or higher than a predetermined temperature, It is possible to judge by In addition, it is conceivable to determine the load operation of the work machine based on information on the lever operation amount, the load pressure, and the like.

According to a fourth aspect of the present invention, there is provided the data processing apparatus according to the first to third aspects of the present invention, wherein the number of neurons to be compressed according to the number of effective real data and the training condition for the compression are set. It further comprises condition setting means (training condition setting unit).
According to a fifth aspect of the present invention, there is provided the data processing device according to the fourth aspect, wherein the training condition setting means includes, as the training conditions, a learning rate, a neighborhood region function, and the number of learning (that is, neuron compression It is characterized by setting calculation conditions that affect the learning result.

According to a sixth aspect of the present invention, in the configuration according to any one of the third to fifth aspects, the neuron compressing means uses the effective real data in the order of the detection time. It is characterized by being compressed into neurons. That is, the permutation of the effective real data used for learning neuron compression is defined based on the detection time.
According to a seventh aspect of the present invention, there is provided a data processing device according to the present invention, wherein, in the configuration according to any one of the first to sixth aspects, a plurality of memories capable of storing the plurality of parameters detected by the parameter detecting means. Means (first storage unit, second storage unit) are further provided.

For example, in the case of a data processing apparatus related to a state determination of a work machine, a first storage unit having a capacity capable of storing operation information for one day of the work machine and a second storage unit having the same capacity as the first storage unit Have it ready. Of course, more storage units may be prepared.
Further, the data processing device of the present invention according to claim 8 is the configuration according to any one of claims 1 to 7, wherein an extraction condition set in advance among the parameters classified into the second parameter group is provided. Parameter extraction means (parameter extraction unit) for extracting a predetermined parameter satisfying the above as an effective parameter, and the basic statistic calculation means calculates a basic statistic of the effective parameter extracted by the parameter extraction means It is characterized by that.

The data processing device of the present invention according to claim 9 is calculated by the neuron obtained by the neuron compression means and the basic statistic calculation means in the configuration according to any one of claims 1 to 8. It is further characterized by further comprising state restoring means (restoring unit) for restoring the state of the object based on the basic statistic.
A remote diagnostic system for a work machine according to a tenth aspect of the present invention is a remote diagnostic system for diagnosing an operating state of a work machine, comprising the data processing device according to the ninth aspect, wherein the parameter detecting means includes A plurality of parameters mounted on a work machine and varying according to the operating state of the work machine are detected, and the parameter group classification means, the neuron compression means, and the basic statistic calculation means are provided as a first information processing apparatus. It is mounted on the work machine as a (first data processing device), and the state restoration device is provided outside the work machine as a second information processing device (second data processing device). The information of the neuron and the basic statistic is acquired from the information processing apparatus by remote communication.

  The data processing method of the present invention according to claim 11 includes a parameter detection step for detecting a plurality of parameters that vary in accordance with the state of the object, and the plurality of parameters detected in the parameter detection step. Multi-dimensional actual data comprising a parameter group classification step for classifying into a first parameter group and a second parameter group in accordance with a predetermined condition based on the amount of variation of each parameter, and parameters classified into the first parameter group Comprising: a neuron compression step for compressing into a neuron by unsupervised learning of a neural network; and a basic statistic calculation step for calculating a basic statistic of each parameter classified into the second parameter group. Yes.

According to a twelfth aspect of the present invention, there is provided the data processing method of the present invention according to the eleventh aspect, wherein, in the parameter group classification step, the plurality of parameters based on a predetermined condition relating to a time variation amount of each of the parameters. It is characterized by classifying.
The data processing method of the present invention according to claim 13 is the configuration according to claim 11 or 12, wherein the data processing method of the present invention is set in advance among a plurality of multidimensional real data comprising parameters classified into the first parameter group. An actual data extraction step for extracting actual data including predetermined parameters satisfying the extraction condition as effective actual data, and in the neuron compression step, the effective actual data extracted in the actual data extraction step is It is characterized by being compressed into neurons by unsupervised learning of the network.

The data processing method of the present invention according to claim 14 is the training according to the configuration according to claims 11 to 13, wherein the number of neurons to be compressed according to the number of effective real data and a training condition related to the compression are set. It further features a condition setting step.
The data processing method of the present invention according to claim 15 is characterized in that, in the configuration according to claim 14, in the training condition setting step, a learning rate, a neighborhood region function, and a learning frequency are set as the training conditions. It is said.

According to a sixteenth aspect of the present invention, in the data processing method of the present invention, in the configuration according to any one of the thirteenth to fifteenth aspects, in the neuron compression step, the effective real data is used in the order of the detection time. It is characterized by being compressed into neurons.
The data processing method of the present invention according to claim 17 is the configuration according to any one of claims 11 to 16, wherein a plurality of memories capable of storing the plurality of parameters detected in the parameter detection step. It is characterized by further comprising a step.

  Further, the data processing method of the present invention according to claim 18 is the configuration according to any one of claims 11 to 17, wherein a preset extraction condition is selected from the parameters classified into the second parameter group. A parameter extracting step for extracting a predetermined parameter satisfying the above as an effective parameter, and calculating a basic statistic of the effective parameter extracted by the parameter extracting means in the basic statistic calculating step. .

The data processing method of the present invention according to claim 19 is calculated by the neuron obtained in the neuron compression step and the basic statistic calculation step in the configuration according to any one of claims 11 to 18. The method further includes a state restoration step of restoring the state of the object based on the basic statistics.
A remote diagnosis method for a work machine according to a twentieth aspect of the present invention is a remote diagnosis method for diagnosing an operating state of a work machine using the data processing method according to any one of claims 12 to 19. In the parameter detection step, a plurality of parameters that vary depending on the operating state of the work machine are detected, and data processing in the parameter group classification step, the neuron compression step, and the basic statistic calculation step is performed. The data processing in the state restoration step is performed on the first information processing apparatus mounted on the work machine, and the information on the neurons and the basic statistics is transmitted from the first information processing apparatus by remote communication. The information is transmitted to a second information processing apparatus provided outside the machine and executed on the second information processing apparatus.

According to the data processing apparatus and the data processing method of the present invention (claims 1 and 10), a plurality of parameters are classified and handled as a first parameter group and a second parameter group, and according to the characteristics of each parameter. Compression can be performed, and compression efficiency can be increased.
In addition, according to the data processing apparatus and the data processing method of the present invention (claims 2 and 12), by using the predetermined condition relating to the time variation amount of the parameter, the parameters suitable for the neuron compression method and the basic statistic are obtained. Parameters suitable for the method can be easily classified.

  In addition, when the amount of time variation of the parameter when the object is normal is relatively small, the state determination using the basic statistics can be performed. That is, the calculation is simple and processing can be performed in a short time. On the other hand, for a subject with a relatively large amount of time variation, the state of the object can be grasped using a neuron compression method, and the state can be accurately determined.

Further, according to the data processing apparatus and the data processing method of the present invention (claims 3 and 13), the reliability of neuron compression can be improved by extracting effective real data from a plurality of real data. .
Further, according to the data processing apparatus and the data processing method of the present invention (claims 4 and 14), the time and accuracy required for compression are arbitrarily adjusted by setting training conditions according to the number of effective real data. be able to.

Further, according to the data processing device and the data processing method of the present invention (claims 5 and 15), by using the learning rate, the neighborhood region function, and the number of learnings as the training conditions, a more approximate performance can be achieved in the compression process. Data characteristics can be retained.
Further, according to the data processing device and the data processing method of the present invention (claims 6 and 16), the effective real data is used in the order of the detection time, so the data related to the latest state at the time of data compression is compressed. be able to. Further, for example, in the case of a hydraulic excavator, it is possible to eliminate effective actual data after startup, particularly before warming up, and the reliability of neuron compression can be further improved.

  Further, according to the data processing device and the data processing method of the present invention (claims 7 and 17), by providing a plurality of storage means, information used for neuron compression can be used properly. For example, it is possible to obtain neurons according to the working day and working content of the object, and data processing using them can be performed. In addition, by sequentially storing the detected plurality of parameters in a plurality of storage units, the work machine can be operated continuously while performing calculations related to neuron compression and calculation of basic statistics.

Further, according to the data processing apparatus and the data processing method of the present invention (claims 8 and 18), the reliability of the calculated basic statistics is improved by extracting effective parameters from a plurality of parameters. Can do.
Further, according to the data processing apparatus and the data processing method of the present invention (claims 9 and 19), the calculation is based on the neurons obtained from the parameters of the first parameter group and the basic statistics of the parameters of the second parameter group. Thus, it is possible to restore the state of the object according to the characteristics of each parameter.

  According to the work machine remote diagnosis system and the work machine remote diagnosis method of the present invention (claims 10 and 20), the parameters according to the operating state of the work machine include the first parameter group, the second parameter group, and the like. Since the data processing is performed by classifying the data into the data, it is possible to compress the amount of information according to the characteristics of each parameter and restore the state of the work machine. In addition, the amount of information transmitted from the first information processing device mounted on the work machine to the second information processing device provided outside the work machine can be compressed, thereby improving the efficiency of remote communication. it can.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
1 to 4 illustrate a data processing apparatus according to an embodiment of the present invention. FIG. 1 is a configuration diagram showing the overall configuration of the data processing apparatus. FIG. 2 relates to data storage in the apparatus. FIG. 3 is a flowchart showing the control contents related to the data compression operation in this apparatus, and FIG. 4 is a three-dimensional graph showing the density distribution of the data restored by this apparatus.

[1. Constitution]
[1-1] Overall Configuration As shown in FIG. 1, the data processing device 15 forms a remote diagnosis system in which the state of a hydraulic excavator (state of normal, abnormal, failure, etc.) is the object of diagnosis. , The data detection device 1 and the first data processing device 10 mounted on the hydraulic excavator, and a maintenance center located outside the hydraulic excavator (outside of the hydraulic excavator provided with the first data processing device 10, For example, it means a business establishment that owns or manages a hydraulic excavator) and a second data processing device 14 for performing information processing.

  The first data processing device 10 obtains a huge amount of detection information (data) reflecting the state of the hydraulic excavator from the data detection device 1, and has a structure (characteristic indicating the state of the hydraulic excavator) inherent in the detection information. An arithmetic process for reducing only the amount of information while maintaining it is performed. That is, in the first data processing device 10, the data size of the detection information is compressed.

  On the other hand, the second data processing device 14 performs arithmetic processing for restoring the information compressed by the first data processing device 10. That is, the detection information is restored in the second data processing device 14.

[1-2] Data detection device The data detection device 1 is a generic name of various sensors corresponding to parameters (detection information, variable elements) related to the hydraulic excavator, and these sensors correspond to the operation of the hydraulic excavator. Each parameter value that fluctuates is detected.

The parameter detected here includes not only data directly detected from the sensor but also data obtained by processing detection information of a certain sensor by calculation or the like and obtaining the value of the corresponding parameter as an estimated value. For example, the current time, total operating time (hour meter) information, operating position information, engine / vehicle error information, accelerator dial (operating dial) information, lever operation amount information, engine speed, engine oil pressure, Boost pressure, fuel consumption, exhaust temperature, power shift pressure, hydraulic pump pressure, load factor (operation rate), engine oil temperature, engine coolant temperature, hydraulic oil temperature, outside air temperature, hydraulic pump case temperature, hydraulic pump drain pressure Etc.
Note that these pieces of detection information are input to the first data processing device 10 as needed together with information on the time detected by the data detection device 1.

[1-3] First Data Processing Device The first data processing device 10 is a small computer including a storage device (ROM, RAM, etc.) and a central processing unit (CPU) incorporating a processing program. Inside the first data processing device 10, there are a preprocessing unit 2, a data collection unit 3, a calculation condition setting unit (calculation condition setting means) 4, a calculation unit 5, a result storage unit 8 and a transmission unit 9 as functional parts. Is provided.

[1-4] Pre-Processing Unit The pre-processing unit 2 converts the detection information input from the data detection device 1 into a digital value format that can be easily calculated inside the first data processing device 10, and An arithmetic process for aligning the number of significant digits is performed.
That is, information on each parameter input from the data detection device 1 includes analog value information and digital value information in accordance with the type of sensor. Further, the accuracy of each detection information is not unified. Therefore, the detection information input from the data detection device 1 is preprocessed by the preprocessing unit 2 once so that the calculation can be easily performed, and then another calculation is performed.

[1-5] Data Collection Unit As shown in FIG. 1, the data collection unit 3 is provided with a first storage unit 3a and a second storage unit 3b (both are one of storage means). Both the first storage unit 3a and the second storage unit 3b have substantially the same storage capacity (memory area), and each of the first storage unit 3a and the second storage unit 3b is a day for a hydraulic excavator. It has a capacity to store the operation information. As described above, by providing the storage units 3a and 3b, it is possible to selectively use the data to be subjected to the compression processing according to the operating date of the excavator and the work content.

In the data collection unit 3 of the present embodiment, the flag F ₁ is set as a condition for determining the proper use of the storage units 3a and 3b. This flag F ₁ is a flag indicating which of the storage units 3a and 3b is the storage unit that is currently recording data (or can record data that is input at that time). Each storage unit 3a, 3b can be switched to F ₁ = 0 or 1 for every operating time that can be stored.

In the present embodiment, the flag F ₁ is switched to F ₁ = 0 or F ₁ = 1 every day (24 hours). Further, when F ₁ = 0, this indicates that the storage unit that is recording data is the first storage unit 3a, and when F ₁ = 1, the storage unit that is recording data is the second storage unit 3b. It is assumed that the state is That is, the data collection unit 3 records information input from the preprocessing unit 2 in the first storage unit 3a when F ₁ = 0, and records it in the second storage unit 3b when F ₁ = 1. To function.

By referring to the flag F _1, the information the storage unit 3a, both the storage unit of the 3b it is possible to know not being recorded the data, for example, recorded in the storage unit during the compression work of the data Can be prevented from being overwritten.
Regarding the switching of the flag F ₁ , the data collection unit 3 of the present embodiment determines that the flag F ₁ is F ₁ = 0 or F by determining whether or not the current time t is a predetermined time t ₁ set in advance. ₁ = 1 is switched. The predetermined time t ₁ is set to midnight.

[1-6] Calculation Condition Setting Unit The calculation condition setting unit (calculation condition setting means) 4 classifies each parameter detected by the data detection device 1 into a first parameter group and a second parameter group. As shown in FIG. 1, extraction conditions for each parameter group are stored therein.
The calculation condition setting unit 4 includes an actual data extraction unit 4a that extracts information to be used for neuron compression calculation from the first parameter group, and a basic statistic calculation from the second parameter group. A parameter extraction unit 4b for extracting information to be provided is provided. The neuron compression referred to here is one of data compression methods, and means the contents of calculation performed by the neuron compression unit 6 described later. The calculation of the basic statistic means the calculation content in the basic statistic calculation unit 7 described later.

  The first parameter group is a general term for parameters that change according to specific work contents, operation types, work modes, and the like by the hydraulic excavator and that have a relatively large amount of time variation. Here, for example, the engine speed, engine oil pressure, boost pressure, fuel consumption, exhaust temperature, power shift pressure, hydraulic pump pressure, load factor (operating rate) and the like are indicated.

  On the other hand, the second parameter group is a general term for parameters that are not easily changed (not easily influenced) by the specific work content, type of operation, work mode, and the like, and that have a relatively small amount of time variation. It is. Here, for example, the engine oil temperature, the engine coolant temperature, the hydraulic oil temperature, the outside air temperature, the hydraulic pump case temperature, the hydraulic pump drain pressure, and the like are indicated. It is known that the temperature information of each part shows a substantially constant value after the warm-up operation when the hydraulic excavator is operating normally.

  That is, in the calculation condition setting unit 4, parameters that have a relatively weak correlation with the elapsed time from the start of operation are extracted as parameters of the first parameter group. On the other hand, as parameters of the second parameter group, parameters that tend to stabilize the fluctuation after a predetermined elapsed time from the start of operation are extracted as parameters of the second parameter group.

[1-7] Extraction of Effective Real Data The actual data extraction unit 4a includes the parameters detected at the same time among the parameters classified into the first parameter group according to the extraction conditions (each parameter detected under the same state). means the, here it sets a multidimensional vector consisting of means) to a respective parameter detected at the same time as the actual data X _s. From this actual data X _s , data used for neuron compression control is extracted and set as effective actual data x _s . It is assumed that the types of parameters included in the actual data X _s extracted by the actual data extraction unit 4a are K types, and the number of effective actual data x _s set here is M. The effective actual data x _s can be generalized and described as the following Expression 1. In the following formulas, symbols representing vectors and matrices are shown in bold (Bold), and symbols meaning scalars are shown in italic.

That is, here, one effective actual data x _s is described as a point coordinate (data point) in the K-dimensional space, and the M effective actual data x _s extracted by the actual data extracting unit 4a is K It is expressed as a set of M points in the dimensional space.
The extraction conditions of the effective actual data x _{s from} the actual data X _s in the actual data extraction unit 4a are as shown below.
(1) Actual data X _s detected during operation at the top three most frequently used accelerator dial positions (2) Not actual data X _s detected during warm-up (idling) of the work machine (3) The actual data X _s detected during the load operation of the work machine Note that the accelerator dial position in the above condition 1 is a general dial type for setting the engine speed of the hydraulic excavator. This means the switch operating position. The above condition 2 is, for example, when the actual data X _s is equal to or higher than a predetermined temperature information such as engine oil temperature, engine coolant temperature, hydraulic oil temperature, outside air temperature, hydraulic pump case temperature, etc. It is determined by whether or not it is detected. The above condition 3 is determined based on whether or not the actual data X _s is detected when the lever operation amount is equal to or greater than a predetermined value set in advance. The determination is made based on the detected value of the pressure switch for detecting the presence or absence of an operation by an operation lever provided in the hydraulic excavator.

Based on such extraction conditions 1 to 3, the actual data extraction unit 4a extracts actual data X _s satisfying these conditions as effective actual data x _s . The extracted effective real data x _s is input to the neuron compression unit 6 of the calculation unit 5.

[1-8] Extraction of Effective Parameter On the other hand, the parameter extraction unit 4b extracts the effective parameter p _s from each parameter P _s classified into the second parameter group based on the following extraction conditions. .
(4) Parameter P _s detected during operation at the top three most frequently used accelerator dial positions (5) Parameter P _s not detected during warm-up (idling) of the work machine (6 ) The work machine is the parameter P _s detected during the load operation. The effective parameter p _s extracted based on these extraction conditions 4 to 6 is input to the basic statistic calculation unit 7 of the calculation unit 5. It has become so.

[1-9] Neuron Compression Unit As shown in FIG. 1, the calculation unit 5 includes a neuron compression unit 6 and a basic statistic calculation unit 7. Each of these functional units will be described below.
The neuron compression unit 6 performs neuron compression control based on the effective actual data x _s extracted by the actual data extraction unit 4a.

First, the neuron compressing unit 6 performs an operation for compressing information on the concept of replacing M pieces of effective real data x _s with N pieces of information particles (N << M), which are significantly smaller in number than this. I do. As a compression method in the neuron compression unit 6, an unsupervised learning method of a neural network is used. Specifically, learning (training) is performed by a known learning method such as a neural gas learning method or a self-organizing map learning method, and N pieces of information particles are obtained as neurons.

For example, a predetermined number of neurons are randomly arranged in the K-dimensional space, and effective real data x _s is input into the K-dimensional space in order of the detection time, and learning of the neurons is repeated. After this learning, weak neurons and idling neurons are deleted, and a neuron set consisting of N neurons is created. Each neuron Thus was created is obtained by compressing the still information quantity characteristic holding the valid actual data x _s (data size), it is described as the same K-dimensional vector effective actual data x _s Yes.

Note that the neuron compressing unit 6 temporarily stores M pieces of effective real data x _s extracted by the actual data extracting unit 4a, and compresses them collectively for each predetermined operating time. For example, thousands of data accumulated by the operation of a hydraulic excavator for one day is used as effective real data x _s and compression into a neuron set is performed every 24 hours.
A training condition setting unit 6 a is provided inside the neuron compression unit 6. The training condition setting unit 6 a is for setting learning conditions for neuron compression in the neuron compression unit 6. Here, according to the number M of effective real data x _s , the number N of neurons to be compressed and the training conditions related to the compression are set. The training conditions in the present embodiment include a learning rate, a neighborhood area function, the number of learnings, and the like.

[1-9-1] Neuron Center For each neuron of the neuron set obtained by the neuron compression control, the neuron compression unit 6 uses the neuron center [c ₁ , c ₂ to obtain information on the position of each neuron in the K-dimensional space. ,..., C _N ]. Since each neuron is configured with K types of parameters, the neuron center c _i is expressed as a K-dimensional vector quantity, and its general expression is as shown in Expression 2 below.

In general, the more the neuron center c _i of a neuron is surrounded by more actual data x _s , the more the neuron can be considered to represent the effective actual data x _s .

[1-9-2] Neuron Weight The neuron compression unit 6 calculates the number of effective real data x _s represented by each neuron as the neuron weight [g ₁ , g ₂ ,..., G _N ]. In other words, the neuron weight g _i indicates how many effective real data x _s of M effective real data x _s are winner neurons. That is, the neuron weights g _i is represented as a scalar quantity. Note that the neuronal weights g _i, it can be said that the indicator of the strength of the neurons.

First, the neuron weight calculation section 1b, for each valid real data x _s, calculates the winner neuron n _s is the shortest neurons Euclidean distance shown in Equation 3 below.

However, D _s (n) is the distance from the effective real data x _s to the i-th neuron. Next, using this information, the neuron compression unit 6 converts the M effective real data x _s into N subsets M Divide into _i (i = 1, 2,..., N). Each subset M _i has effective real data x _s with the same i-th neuron as a winner neuron. When the number of effective actual data x _s included in each subset M _i is expressed as m _i (i = 1, 2,..., N), the conditions shown in the following expressions 4 and 5 are satisfied.

That is, the standardized weight g _i of the neuron is calculated by Expression 6 shown below.

Each subset M _i (i = 1, 2,..., N) forms a unique subspace in the K-dimensional space, and has a shape similar to a so-called “Voronoi polygon”. A Voronoi polygon means a polygon having a complicated shape formed by vertical bisectors of adjacent points when a large number of points exist in a plane.

[1-9-3] Neuron Width Further, the neuron compression unit 6 determines the average distance to each effective real data x _s existing in the subset M _i for each neuron as the neuron width [w ₁ , w ₂ ,. , W _N ]. This is because there is a portion where the effective real data x _s does not partially exist in the region of each subset M _i , and the size of the region surrounded by the Voronoi polygon necessarily reflects the distribution state of the effective real data x _s . It's not a thing. Here, the neuron width w _i (i = 1, 2,..., N) is treated as an index indicating the distribution state of the effective real data x _s . The neuron width w _i is expressed as a K-dimensional vector quantity, and the general formula thereof is as shown in the following formulas 7 and 8.

Note that information on the neuron set set and calculated in the neuron compression unit 6, that is, information on the neuron center c _i , neuron weight g _i, and neuron width w _i is input to the result storage unit 8. It has become.

[1-10] Basic statistic calculation unit On the other hand, the basic statistic calculation unit 7 calculates the basic statistic of each effective parameter p _s based on the effective parameter p _s extracted by the parameter extraction unit 4b. It is.

The basic statistic of the effective parameter p _s calculated by the basic statistic calculating unit 7 refers to a statistic representing the basic characteristics of each effective parameter p _s , and here, the number of data and the average value of the data , Representative values such as mode value, maximum value and minimum value, and the degree of dispersion such as standard deviation.
The basic statistic calculated by the basic statistic calculation unit 7 is input to the result storage unit 8 in the same manner as the neuron set information.

In the calculation unit 5, during the implementation of operations neuronal compression as described above, the flag F ₂ is adapted to be set. The flag F ₂ is a flag indicating whether or not neuron compression is being calculated. In this embodiment, F ₂ = 1 is set during neuron compression, and compression calculation is performed. Is set so that F ₂ = 0.

[1-11] Result Storage Unit / Transmission Unit The result storage unit 8 is a storage device that stores information on the neuron set and basic statistics calculated by the calculation unit 5. In addition, here, together with the compressed information such as the above neuron center c _i , neuron weight g _i , neuron width w _i, and basic statistics, information on the total operating time (integrated operating time, hour meter) of the hydraulic excavator, operating position Information, engine / vehicle error information is saved.

  The transmission unit 9 is for wirelessly transmitting each of the above information including information on neuron sets and basic statistic information stored in the result storage unit 8 to the outside of the excavator. Here, the lowest-cost telecommunications network (such as a telephone network or satellite communication) is used, and each piece of information is transmitted to the second data processing device 14 provided in the maintenance center. . The communication frequency in the transmission unit 9 corresponds to the compression frequency in the neuron compression unit 6, and is transmitted once a day after the operation of the hydraulic excavator, for example.

  Each piece of information not transmitted by the transmission unit 9 is stored in the result storage unit 8 for a certain period. Since most of the information stored here is information that has undergone neuron compression and calculation of basic statistics, the data size is more compact than the information stored in the data collection unit 3.

[1-12] Second Data Processing Device Like the first data processing device 10, the second data processing device 14 includes a storage device (ROM, RAM, etc.) and a central processing unit (CPU) incorporating a processing program. The computer is configured to perform arithmetic processing for restoring information transmitted from the first data processing device 10. The second data processing device 14 includes a receiving unit 11, a restoring unit 12, and a data output device 13 as functional parts. First, the reception unit 11 functions to receive information transmitted from the transmission unit 9 of the first data processing apparatus 10 and input the information to the restoration unit 12.

[1-13] Restoration Unit On the other hand, the restoration unit 12 performs a calculation based on the information of the above neuron set (that is, the neuron center c _i , the neuron weight g _i , and the neuron width w _i ) and compresses the neuron An operation for restoring the structure of the original effective real data x _s from the set is performed.
The restoration of information here is an idea different from, for example, data expansion or decompression (decompression, unzipping) in the software application field. The main difference is not to completely restore the original information, but to extract the correlation between the parameters of the original data based on the compressed information and to restore the outline of the density distribution of the data It is a point.

First, the restoration unit 12 performs a restoration calculation of the density distribution of the effective real data x _s from the neuron information compressed by the calculation unit 5. First, a three-dimensional graph as shown in FIG. 4 is created based on the three pieces of information of the neuron center c _i , neuron weight g _i, and neuron width w _i input from the first data processing device 10. FIG. 4 is a graph showing the density distribution by extracting only two parameters of the boost pressure and the engine speed from the information included in the neuron set.

On this graph, each neuron is shown as a peak (mountain), and a plurality of neurons are visualized in FIG. The position of the peak (mountain) on the graph corresponds to the value of the neuron center c _i . Further, the height of the peak is determined corresponding to the value of the neuron weight g _i , and the slope of the peak (gradient) is set corresponding to the value of the neuron width w _i . For example, a neuron with a larger neuron weight g _i is drawn with a higher peak, and a neuron with a larger neuron width w _i is drawn with a gentle slope.

Also, here, two parameters included in the neuron set are extracted and graphed, but when one parameter is extracted and graphed, the density distribution can be drawn as a two-dimensional graph. Such graphing processing is performed for all parameters included in the neuron set.
The restoration unit 12 outputs the information on the density distribution of the neurons restored as described above and the information on the basic statistics transmitted from the first data processing device 10 to the data output device 13 for display. It is supposed to be displayed.

[2. flowchart]
2 and 3 are flowcharts implemented in the data processing device 15 according to the present embodiment. Both of these flowcharts are flows executed on-board inside the first data processing apparatus 10.

[2-1] Storage Flow The flow of FIG. 2 shows the control contents related to storage of each parameter value detected by the data detection device 1.

First, in step A10, each parameter value detected by the data detection device 1 is input to the first data processing device 10. Each parameter input here is accompanied by information on the time detected by the data detection device 1.
In the subsequent step A20, the preprocessing unit 2 converts each parameter value detected by the data detection device 1 into a digital value format, and performs a process of aligning the number of significant digits of each parameter, and then proceeds to step A30. move on.

In step A30, the data collection unit 3 determines whether or not the flag F ₁ is F ₁ = 0. If F ₁ = 0, the process proceeds to step A40. If F ₁ ≠ 0 (that is, F ₁ = 1), the process proceeds to step A50.
In step A40, in the data collection unit 3, the information of each parameter input from the preprocessing unit 2 is recorded in the first storage unit 3a, and the process proceeds to step A60. On the other hand, in step A50, information on each parameter is recorded in the second storage unit 3b, and the process proceeds to step A60. In other words, the determination of the flag F ₁ makes it possible to grasp which of the storage units 3a and 3b is not recording data.

In step A60, the calculation unit 5 determines whether or not the flag F ₂ is F ₂ = 1. That is, it is determined at this point whether or not neuron compression is being performed. If F ₂ = 1, the operation proceeds to the neuron compression and basic statistics calculation flow. On the other hand, if F ₂ ≠ 1 (ie, F ₂ = 0), the process proceeds to step A70.
In step A70, the data acquisition unit 3, the current time t whether a predetermined time t ₁ is determined. If t ≠ t ₁ in this step, that is, if the time is not midnight, this flow is terminated as it is, and the control from step A10 is repeated at a predetermined cycle. In this case, since the value of the flag F ₁ is not switched, the determination result in step A30 does not change, and either control in step A40 or step A50 is repeatedly performed.

On the other hand, when t = t ₁ in Step A70, that is, when the time is midnight, the control of Step A80 to Step A100 is performed, and the value of the flag F ₁ is switched. Specifically, in step A80, it is determined whether or not the flag F ₁ is F ₁ = 0. If F ₁ = 0, the process proceeds to step A90 and F ₁ = 1 is set. Conversely, if F ₁ ≠ 0 (ie, F ₁ = 1) in step A80, the process proceeds to step A100 and F ₁ = 0 is set.

In addition, after the flag setting in step A90 and step A100, it proceeds to the calculation flow of neuron compression and basic statistics.
As described above, each parameter detected by the data detection device 1 is recorded in either the first storage unit 3a or the second storage unit 3b unless the date is straddled (unless the time is midnight). to continue. When the date is straddled (when the time is midnight), the storage unit in which each parameter is recorded is switched.

[2-2] Calculation Flow of Neuron Compression and Basic Statistics The flow of FIG. 3 shows the details of specific calculation processing using each parameter value stored in the previous flow.
First, in step B10, it is determined whether or not the flag F ₁ is F ₁ = 0. In this step, a determination is made to select a storage unit in which information to be calculated in the following steps is stored. That is, if it is determined in this step that F ₁ = 0, the information input from the preprocessing unit 2 should be recorded in the first storage unit 3a at that time, so the second storage unit 3b Each parameter stored in is read and used as a calculation target. Conversely, if it is determined in this step that F ₁ = 1, since the information input from the preprocessing unit 2 should be recorded in the second storage unit 3b at that time, the first storage unit Each parameter stored in 3a is read and used as a calculation target.

In the subsequent step B40, the calculation condition setting unit 4 classifies the parameters read in the previous step based on a predetermined condition relating to the amount of time variation.
That is, among the read parameters, parameters such as engine speed, engine oil pressure, boost pressure, fuel consumption, exhaust temperature, power shift pressure, hydraulic pump pressure, load factor, etc., with a relatively large amount of time fluctuation are included. The first parameter group. Further, parameters such as engine oil temperature, engine coolant temperature, hydraulic oil temperature, outside air temperature, hydraulic pump case temperature, hydraulic pump drain pressure, and the like, which have a relatively small amount of time fluctuation, are classified into the second parameter group.

In the subsequent step B50, the actual data extraction unit 4a of the calculation condition setting unit 4 sets actual data X _s composed of parameters detected at the same time among the parameters classified into the first parameter group, further, the effective actual data x _s to be used for neurons compression control from the real data X _s is extracted. The conditions of the effective actual data x _s extracted here are (1) those detected at the time of operation at the three most frequently used accelerator dial positions, and (2) engine oil temperature and engine coolant temperature. Is detected when the temperature is equal to or higher than a predetermined temperature set in advance, and (3) is detected when an operation with the operation lever is performed. By such conditioning, the information of the actual data X _s detected in an unstable and transient state such as during the warm-up operation of the excavator is excluded, and the reliability of the data set subjected to compression is ensured. .

In the subsequent step B60, the neuron compression unit 6 uses an unsupervised learning method of the neural network to start calculation of a plurality of neurons (neuron sets). Conditions of learning in this step, in the training condition setting unit 6a, is set according to the number M of active real data x _s. In this embodiment, the learning rate, the neighborhood region function, and the number of learnings are set as the training conditions. By the compression operation in this step, the information amount is compressed in a state where the characteristic of the effective real data x _s is held in the neuron.

In this step, the neuron center c _i , neuron weight g _i, and neuron width w _i are calculated as compressed neuron set information. These pieces of information are input to the result storage unit 8. Note that the neuron compression operation in this step is processed as a subflow independent of this flow.
On the other hand, in the subsequent step B70, the parameter extraction unit 4b of the calculation condition setting unit 4 extracts the parameters P _s classified into the second parameter group for actual calculation as effective parameters p _s. The The extraction conditions of the effective parameter p _s extracted here are the same as the extraction conditions of the effective actual data x _s in step B50, and (1) detected during operation at the three most frequently used accelerator dial positions. (2) Detected when the engine oil temperature and the engine coolant temperature are equal to or higher than a predetermined temperature set in advance. (3) Detected when the operation lever is operated. It is a thing. By such conditioning, information on the parameter P _s detected in an unstable and transitional state such as during the warm-up operation of the hydraulic excavator is excluded, and the reliability of the parameter provided for the calculation is ensured.

In step B80, the basic statistic calculation unit 7 starts calculating the basic statistic of the extracted effective parameter p _s . In the present embodiment, the number of data of the extracted effective parameter p _s , the average value, the mode value, the maximum value, the minimum value, and the standard deviation of the data are calculated. These basic statistics are input to the result storage unit 8. Note that the calculation of basic statistics in this step is also processed as a subflow independent of this flow.

In subsequent step B90, it is determined whether or not the calculation of the neurons and the basic statistics in the calculation unit 5 has been completed for all the effective real data x _s and the effective parameters p _s . If it is determined that the process has been completed, the process proceeds to step B100. If it is determined that the process has not been completed, the process proceeds to step B110.
In step B100, the flag F ₂ is set to F ₂ = 0 and this flow is ended. On the other hand, in step B110, the flag F ₂ is set to F ₂ = 1 and this flow is ended. That is, when the calculation of the neuron and the basic statistics is not completed, the flag F ₂ remains in the state of F ₂ = 1. Therefore, when the storage flow shown in FIG. Proceed to the calculation flow. Further, once the calculation of the neuron and the basic statistics is completed, the flag F ₂ is set to F ₂ = 0, so that the calculation flow is not executed until the time reaches midnight.

Note that the neuron and basic statistic information input to the result storage unit 8 by this calculation flow is transmitted to the second data processing device 14 of the maintenance center by the transmission unit 9 after the calculation ends. .
On the other hand, the restoration unit 12 of the second data processing device 14 restores the data density distribution based on the transmitted neuron information. The restored density distribution and basic statistics are displayed on the data output device 13, for example, as shown in FIG.

[3. effect]
The data processing device 15 according to an embodiment of the present invention has the following effects.
[3-1] Advantages of the Neuron Compression Method First, the neuron compression unit 6 of the first data processing apparatus 10 has a configuration for compressing the parameters classified into the first parameter group into neurons. That is, for example, parameters having a large time change, such as engine speed and hydraulic pump pressure, in other words, parameters having a weak correlation with the operating time are compressed into neurons.

  Since the parameters classified into these first parameter groups vary greatly depending on the work content and type of operation of the excavator, for example, even if a single parameter is extracted and its average value or variation is observed, It is difficult to specifically determine the state of the excavator. However, by compressing these parameters into a multi-dimensional neuron, it is possible to compress and restore approximately while maintaining the correlation structure between the parameters. Therefore, a change in the state of the hydraulic excavator can be detected by detecting such a change in the correlation structure, and can be used, for example, for determining a failure of the hydraulic excavator.

  Further, as shown in FIG. 4, if the correlation between the two parameters is graphed, it is possible to visually grasp the operating state of the hydraulic excavator. That is, for example, by preparing a correlation graph between arbitrary parameters during normal operation of the hydraulic excavator, it is easy to diagnose the operating state of the hydraulic excavator using that as a criterion.

[3-2] Advantages of the Basic Statistics Method On the other hand, the basic statistic calculator 7 of the first data processing apparatus 10 has a configuration for calculating the basic statistic of the parameters classified into the second parameter group. Yes. That is, for example, parameters with small time changes such as engine oil temperature and engine coolant temperature, in other words, basic statistics of parameters that take substantially constant values as long as the hydraulic excavator is operating normally once warmed up. The amount is being calculated.

  Since the parameters classified into these second parameter groups have a small amount of change in the detected data, the approximate characteristics can be grasped by observing their average values and variations. Therefore, the state of the excavator can be determined by comparing the basic statistics of these parameters with a predetermined threshold value set in advance. For example, when a deviation between the basic statistics and the predetermined threshold value is detected, It can be determined that there is a sign of failure.

[3-3] Advantages of the Two Techniques The data processing device 15 further includes a calculation condition setting unit 4 that classifies each parameter detected by the data detection device 1 into a first parameter group and a second parameter group. ing. In other words, instead of compressing all detected information into neurons, it classifies the detected information into information used for neuron compression and information used for simpler arithmetic processing, and only the former information is sent to neurons. And it comes to compress. As a result, calculation processing according to the characteristics of each parameter is performed, and the amount of information can be further reduced while maintaining the structure (characteristic indicating the state of the hydraulic excavator) inherent in the detection information with a simple structure. It becomes possible. That is, the state of the hydraulic excavator can be analyzed accurately and efficiently.

  Of the two methods described above, the calculation of the basic statistic can be processed easily and in a short time compared to the calculation of neuron compression, and the data size of the calculation result is small. Therefore, according to the data processing device 15, it is possible to grasp the operating characteristics of the hydraulic excavator more efficiently than when all the parameters are compressed into neurons. In addition, the amount of communication from the first data processing device 10 to the second data processing device 14 is reduced, and the communication cost can be reduced.

  In the data processing device 15, the first parameter group and the second parameter group are classified according to the amount of time variation of each parameter. Parameters suitable for the method can be easily classified, and the state of the excavator can be easily determined.

[3-4] Advantages of Extracting Effective Real Data and Effective Parameters The data processing apparatus 15 includes the operation condition setting unit 4 including an actual data extracting unit 4a and a parameter extracting unit 4b. That is, information relating to neuron compression and calculation of basic statistics is extracted using information when the excavator is performing load operation as effective data. Thereby, for example, information at the time of idling of the excavator can be excluded, and the reliability of the calculation result in the calculation unit 6 can be improved.

  In addition, since the actual data extraction unit 4a and the parameter extraction unit 4b do not refer to information at the time of warm-up operation of the hydraulic excavator, it is possible to exclude transient data such as pressure not being stable, The calculation accuracy can be further increased.

[3-5] In the training condition setting unit 6a of the advantages the present data processing apparatus 15 related to the training conditions, effective according to the number M of the real data x _s, training conditions according to the number N and the compression of the compressible neurons is set It is like that. Thereby, the time required for neuron compression and the compression accuracy can be arbitrarily adjusted.

Further, during the compression calculation in the neuron compression unit 6, the effective real data x _s is learned in the order of the detection time, so that data related to the latest state at the time of data compression can be compressed. . In particular, considering the general usage of hydraulic excavators, learning from the latest effective actual data x _s makes it easier to eliminate unstable effective actual data x _s before warm-up, and the reliability of neuron compression The property can be further improved.
The number M of effective real data x _{s used} for neuron compression is preferably about 450 to 6000.

[3-6] Advantages of Multiple Storage Units In the data processing device 15, the data collection unit 3 includes two storage units, a first storage unit 3a and a second storage unit 3b. Information on each parameter to be input is recorded alternately in these storage units every day. As a result, the data to be subjected to the compression process can be properly used according to the working day of the excavator and the work content.

For example, in this embodiment, the value of the flag F ₁ is switched when the time is midnight, and the calculation flow is started. However, the hydraulic excavator operates at midnight across midnight. Even if it is, it can memorize | store the information at the time of the detected operation in the data collection part 3 without delay, implementing the calculation in the calculating part 5. FIG. As described above, the plurality of detected parameters are sequentially stored in the plurality of storage units 3a and 3b, so that the hydraulic excavator can be operated continuously while performing calculations related to neuron compression and calculation of basic statistics. .

[4. Others]
Although one embodiment of the present invention has been described above, the present invention is not limited to this embodiment, and various modifications can be made without departing from the spirit of the present invention.
For example, in the above-described embodiment, the data processing device 15 as a remote diagnosis system including the first data processing device 10 and the second data device 14 is shown, but the second data processing device 14 is omitted. It is also possible to do. That is, when the neuron set and basic statistics obtained by the first data processing device 10 are used as they are without performing arithmetic processing, the operation state is diagnosed at least as in the first data processing device 10. Any configuration including a simple data processing device may be used.

In the above-described embodiment, as shown in FIG. 4, the density distribution restored on the first data processing device 10 is displayed on the data output device 13 as a three-dimensional graph. However, it is conceivable to apply various known methods for the output method of the analysis result.
For example, the actual data during normal operation of the hydraulic excavator is detected in advance as normal actual data, the normal actual data is neuron-compressed to obtain a normal neuron set, and the neuron set calculated by the neuron compression unit 6 is obtained. The operation state of the hydraulic excavator may be determined by comparing the normal neuron set with the above-described normal neuron set.

  In addition, basic statistic information may be superimposed on a neuron density distribution graph. For example, the basic statistics of the parameters of the second parameter group detected at the same time as the effective actual data represented by each neuron are displayed on the graph, or only those neurons whose basic statistics are greater than or equal to a preset threshold To highlight. By combining the diagnostic method using information on the neuron set and the diagnostic method using basic statistics, the calculation result in the neuron compression unit 6 and the calculation result in the basic statistical value calculation unit 7 are used. Correspondence becomes clear and it becomes possible to determine the state of the excavator more accurately.

Further, in the above-described embodiment, each parameter detected by the data detection device 1 is classified into the first parameter group and the second parameter group, but such a configuration is not essential. For example, the calculation condition setting unit 4 may classify all the parameters detected by the data detection device 1 into the first parameter group. In this case, all the information obtained from the data detection device 1 becomes the actual data X _s and becomes the extraction target of the effective actual data x _s in the actual data extraction unit 4a.

Further, in the above-described embodiment, in the neuron compression control in the neuron compression unit 6, the effective real data x _s is learned in the order of the detection time, but other calculation method variations are also conceivable. . For example, learning may be performed in order from the oldest detection time, or only effective actual data x _s posted within a predetermined time as a reference set in advance may be used for learning of neurons.

In the above-described embodiment, three conditions are defined as conditions for extracting the effective actual data x _s from the actual data X _s . However, only one of these conditions may be provided. Alternatively, a configuration in which extraction conditions are further added may be used. The same applies to the extraction conditions for the effective parameter p _s . In the calculation of basic statistics, the larger the population (number of effective parameters p _s ), the more reliable the results (average value, mode value, maximum value, minimum value, standard deviation, etc.). Therefore, from the viewpoint of the reliability of the results, it is important that the extraction conditions are not too strict.

  In the above-described embodiment, the result storage unit 8 of the first data processing apparatus 10 stores information that has undergone neuron compression and calculation of basic statistics for a certain period. It is also possible to perform another calculation process using the information. Similarly, the excavator operation information recorded in the data collection unit 3 can be used not only for the calculation process in the calculation unit 5 but also for other calculations. In this case, for example, by increasing the storage capacity of the data collection unit 3, the detection information input from the data detection device 1 can be used as statistical material.

  In the present embodiment, a hydraulic excavator is taken as an example of a diagnosis target, but the target is not limited to this, for example, vehicles such as trucks and buses, ships, and industrial machines. The present invention can be widely applied to the determination of the quality of operations of various machinery, and the like, and can also be applied to the determination of the quality of life forms such as animals and plants and microorganisms, and estimation of changes in celestial bodies such as weather or the earth.

It is a block diagram which shows the whole structure of the data processor concerning one Embodiment of this invention. It is a flowchart which shows the control content which concerns on the memory | storage of data in this data processor. It is a flowchart which shows the control content which concerns on the compression calculation of the data in this data processor. It is a three-dimensional graph which shows the density distribution decompress | restored with this data processor.

Explanation of symbols

1 Data detection device (parameter detection means)
2 Pre-processing unit 3 Data collection unit 3a First storage unit (one storage means)
3b Second storage unit (one of storage means)
4 Calculation condition setting section (Calculation condition setting means)
4a Actual data extraction unit (actual data extraction means)
4b Parameter extraction unit (parameter extraction means)
5 arithmetic unit 6 neuron compression unit (neuron compression means)
6a Training condition setting section (training condition setting means)
7 Basic statistic calculator (basic statistic calculator)
8 result storing unit 9 transmitting unit 10 first data processing device 11 receiving unit 12 restoring unit (state restoring unit)
13 Data Output Device 14 Second Data Processing Device 15 Data Processing Device

Claims (20)

Parameter detecting means for detecting a plurality of parameters that vary depending on the state of the object;
Parameter group classification means for classifying the plurality of parameters detected by the parameter detection means into a first parameter group and a second parameter group;
Neuron compression means for compressing multidimensional real data consisting of parameters classified into the first parameter group into neurons by unsupervised learning of a neural network;
A data processing apparatus comprising basic statistic calculation means for calculating basic statistic of each parameter classified in the second parameter group. 2. The data processing apparatus according to claim 1, wherein the parameter group classification means classifies the plurality of parameters based on a predetermined condition relating to a time variation amount of each of the parameters. Real data extraction means for extracting, as effective real data, real data including a predetermined parameter satisfying a preset extraction condition among a plurality of multidimensional real data including parameters classified into the first parameter group In addition to providing
The data processing apparatus according to claim 1 or 2, wherein the neuron compression means compresses the effective real data extracted by the real data extraction means into neurons by unsupervised learning of a neural network. 4. The data processing apparatus according to claim 1, further comprising training condition setting means for setting the number of neurons to be compressed according to the number of effective real data and a training condition relating to the compression. 5. The data processing apparatus according to claim 4, wherein the training condition setting means sets a learning rate, a neighborhood region function, and the number of learnings as the training condition. The data processing apparatus according to any one of claims 3 to 5, wherein the neuron compression means compresses the effective real data into neurons using the detection data in order from the latest detection time. The data processing apparatus according to claim 1, further comprising a plurality of storage units capable of storing the plurality of parameters detected by the parameter detection unit. Among the parameters classified into the second parameter group, the apparatus further comprises parameter extraction means for extracting a predetermined parameter that satisfies a preset extraction condition as an effective parameter,
The data processing apparatus according to claim 1, wherein the basic statistic calculation unit calculates a basic statistic of the effective parameter extracted by the parameter extraction unit. Further comprising state restoration means for restoring the state of the object based on the neurons obtained by the neuron compression means and the basic statistics calculated by the basic statistics calculation means; The data processing device according to claim 1. A remote diagnosis system comprising the data processing device according to claim 9 and diagnosing an operating state of a work machine,
The parameter detection means is mounted on the work machine, detects a plurality of parameters that vary according to the operating state of the work machine,
The parameter group classification means, the neuron compression means, and the basic statistic calculation means are mounted on the work machine as a first information processing apparatus,
The state restoration device is provided outside the work machine as a second information processing device, and acquires information on the neuron and the basic statistics from the first information processing device by remote communication. A remote diagnosis system for work machines. A parameter detection step for detecting a plurality of parameters that vary according to the state of the object;
A parameter group classification step for classifying the plurality of parameters detected in the parameter detection step into a first parameter group and a second parameter group according to a predetermined condition based on a variation amount of each of the plurality of parameters;
A neuron compression step of compressing multidimensional real data composed of parameters classified into the first parameter group into neurons by unsupervised learning of a neural network;
A data processing method comprising: a basic statistic calculation step of calculating a basic statistic of each parameter classified into the second parameter group. 12. The data processing method according to claim 11, wherein, in the parameter group classification step, the plurality of parameters are classified based on a predetermined condition relating to a time variation amount of each of the parameters. An actual data extraction step for extracting, as effective real data, real data including a predetermined parameter satisfying a preset extraction condition among a plurality of multidimensional real data consisting of parameters classified into the first parameter group In addition to preparing
13. The data processing method according to claim 11 or 12, wherein in the neuron compression step, the effective real data extracted in the real data extraction step is compressed into neurons by unsupervised learning of a neural network. The data processing method according to claim 11, further comprising a training condition setting step of setting a training condition related to the number of neurons to be compressed and the compression according to the number of effective real data. 15. The data processing method according to claim 14, wherein, in the training condition setting step, a learning rate, a neighborhood region function, and a learning frequency are set as the training conditions. The data processing method according to any one of claims 13 to 15, wherein, in the neuron compression step, the effective real data is compressed into neurons using the detection data in the order of detection time. The data processing method according to any one of claims 11 to 16, further comprising a plurality of storage steps capable of storing the plurality of parameters detected in the parameter detection step. A parameter extracting step of extracting, as an effective parameter, a predetermined parameter that satisfies a preset extraction condition among the parameters classified into the second parameter group;
18. The data processing method according to claim 11, wherein, in the basic statistic calculation step, a basic statistic of the effective parameter extracted by the parameter extracting unit is calculated. The method further comprises a state restoration step for restoring the state of the object based on the neuron obtained in the neuron compression step and the basic statistic calculated in the basic statistic calculation step. The data processing method according to any one of claims 11 to 18. A remote diagnosis method for diagnosing an operating state of a work machine using the data processing method according to claim 19,
In the parameter detection step, a plurality of parameters that vary depending on the operating state of the work machine are detected,
The data processing in the parameter group classification step, the neuron compression step, and the basic statistic calculation step is performed on the first information processing apparatus mounted on the work machine,
The data processing in the state restoration step is performed by transmitting information on the neurons and the basic statistics from the first information processing device to a second information processing device provided outside the work machine by remote communication, A remote diagnosis method for a work machine, which is performed on a second information processing apparatus.

Images