JP6169473B2 - Work machine management device and work machine abnormality determination method - Google Patents

Description

  The present invention relates to a management apparatus and an abnormality determination method for determining abnormality of a work machine based on a detected value of a physical quantity acquired from the work machine.

  Work machines such as excavators are used in various construction sites, civil engineering sites, and the like, and when a failure occurs, quick failure repair is required. Diagnosis systems that detect abnormalities based on various parameters that vary depending on the state of the work machine have been developed (Patent Documents 1 and 2). For example, an abnormality is detected based on a plurality of parameters such as engine speed and hydraulic pressure. As an example, time integrated values of various parameters collected from the work machine are used. By performing time integration, the influence of noise can be eliminated.

JP 2006-53818 A JP 2007-257366 A

  When time integration of the detected various parameters is performed, even if an abnormal fluctuation occurs in a short time, the fluctuation is not detected. An object of the present invention is to provide a management device and an abnormality determination method capable of detecting an abnormal change in a short time and determining an abnormality of a work machine such as a shovel.

According to one aspect of the invention,
A storage device for storing a plurality of reference data representing time waveforms at the time of normal operation of a plurality of operation variables measured by sensors mounted on the work machine and depending on an operating state of the work machine;
A processing device,
The processor is
The time waveform of the operation variable obtained by measuring the plurality of operation variables of the work machine to be evaluated is divided into a plurality of sections on the time axis, and the time of the plurality of operation variables for each section. Obtain verification data including values calculated based on the waveform,
For each of the sections, by determining the discretized feature value based on the value of the operation variable of the verification data, the discretization verification data including the time series of the feature value is obtained,
A work machine management device is provided that determines whether or not the work machine to be evaluated is abnormal based on the reference data and the discretization verification data.

According to another aspect of the invention,
The time waveform of the operating variable , which is obtained by measurement with the sensor mounted on the work machine to be evaluated and depends on the operating state of the work machine, is divided into multiple sections on the time axis, and is obtained for each section. Obtaining verification data including values of a plurality of the operating variables;
Determining discretized verification data consisting of a time series of the feature values by determining a discretized feature value for each section based on the value of the operation variable of the verification data;
Whether the work machine to be evaluated is abnormal based on a plurality of reference data representing a time waveform of the operating variable at a normal time obtained during normal operation of the work machine and the discretized verification data. There is provided an abnormality determination method for a work machine having a step of determining whether or not.

  To determine whether a work machine is abnormal using discretized verification data consisting of time series of feature values of operating variables, perform high-accuracy abnormality determination in consideration of abnormal variations in a short time Can do.

FIG. 1 is a block diagram of a work machine management device according to an embodiment and a side view of a work machine that is a management target. FIG. 2 is a graph showing an example of time waveforms of the operation variables α, β, γ,... When the boom is lowered and the boom is raised repeatedly. FIG. 3 is a chart showing an example of a plurality of reference data. FIG. 4 is a flowchart of an abnormality determination process executed by the management apparatus according to the embodiment. FIG. 5 is a chart showing an example of the verification data. FIG. 6 is a flowchart showing the process of step SA3 of the flowchart shown in FIG. FIG. 7 shows an example of standardized reference data obtained by standardizing the operating variables A to I by excluding the sections D5 to D7 from the population based on the reference data shown in FIG. It is a chart. FIG. 8A is a chart showing an example of the discretization reference data, and FIG. 8B is a chart showing an example of the discretization verification data. FIG. 9 is a flowchart showing the process of step SA4 of the flowchart shown in FIG. FIG. 10 is a state transition diagram of the state transition model. FIG. 11 is a chart showing a plurality of state sequences generated for discretized reference data. FIG. 12 is a chart showing a plurality of state sequences generated for discretized verification data. It is a graph which shows the score calculated | required about each of reference data RD and verification data ED.

  FIG. 1 shows a block diagram of a work machine management apparatus 10 according to an embodiment and a side view of a work machine 18 to be managed. The management device 10 includes an input / output device 11, a processing device 12, and a storage device 13. A sensor 20, a control device 21, and a communication device 22 are mounted on the work machine 18. The sensor 20 measures various operating variables of the work machine 18. The measurement result of the sensor 20 is input to the control device 21. The communication device 22 transmits / receives data to / from the management device 10 via the communication line 19. The control device 21 transmits measured values of various operating variables to the management device 10 by controlling the communication device 22.

  The input / output device 11 of the management device 10 includes, for example, a communication device, an image display device, a keyboard, a pointing device, and the like. The measured values of the operating variables measured by various sensors 20 mounted on the work machine 18 are input to the input / output device 11 via the communication line 19. Based on the measured value input to the input / output device 11, the processing device 12 performs an abnormality determination process. In the abnormality determination process, various data stored in the storage device 13 are used. The abnormality determination result is output to the image display device of the input / output device 11.

  Storage areas for reference data RD, standardized reference data RDS, discretized reference data RDD, verification waveform EW, verification data ED, standardized verification data EDS, and discretized verification data EDD are secured in the storage device 13. The reference data RD is stored in the storage device 13 in advance. The reference data RD is prepared by measuring a plurality of operating variables depending on the operating state of the work machine during normal operation of the work machine. The operating variables include, for example, engine torque, engine speed, cooling water temperature, intake air temperature, boost pressure, boost temperature, oil temperature, and the like. These operating variables are measured in the work machine 18 with a time step of, for example, 50 ms. The measured value of the operating variable is transmitted from the work machine 18 to the management device 10.

  With reference to FIG.2 and FIG.3, the acquisition method of reference data RD and the structure of reference data RD are demonstrated.

  In FIG. 2, the operation variables α, β, when the boom is lowered (hereinafter referred to as “first state”) and the boom is raised (hereinafter referred to as “second state”) are repeated. An example of a time waveform of γ,... A period in which the operation variable α is relatively small and a period in which the operation variable α is relatively large are alternately repeated. The operating variable β varies with substantially the same cycle as the cycle of variation of the operating variable α. The operation variable γ varies independently of the variation cycle of the operation variable α. A period during which the operation variable α is relatively small and a period during which the operation variable α is relatively large correspond to the first state and the second state, respectively.

  Based on the time waveform of the operation variable α, waveforms of a plurality of periods during which the same operation is performed are cut out from the time waveforms of the plurality of operation variables. The cut out waveform is referred to as a “partial waveform”. In the present embodiment, a waveform for a certain period (for example, 10 seconds) including the time when the waveform of the operation variable α rises is cut out from the reference waveform of each of the plurality of operation variables. The cut partial waveform is divided into a plurality of sections, for example, ten sections D1 to D10 on the time axis. For each section, statistics of operating variables, for example, average value and standard deviation are obtained. The average value and standard deviation of the operation variables for each section can also be referred to as operation variables representing the operating state of the work machine. Thus, not only the physical quantity directly measured by the sensor 20 mounted on the work machine 18 but also the statistical quantity obtained by processing the measured value of the operating variable using the statistical method is also called the operating variable. it can.

  When each of the 10-second partial waveforms cut out from the reference waveform of the operating variable is divided into 10 sections D1 to D10, the length of one section is 1 second. When the measurement cycle of the operating variable is 50 ms, 20 measurement values are included in one section. From these 20 measured values, the average value and standard deviation of the section of the operating variable are calculated. Reference data RD is obtained from each of the extracted partial waveforms. FIG. 2 shows an example in which partial waveforms serving as the basis of seven reference data RD1 to RD7 are cut out from the waveforms of the operating variables α, β, γ,.

  FIG. 3 shows an example of a plurality of reference data RD. Each of the reference data RD1, RD2, RD3,... Includes a plurality of operation variables A to I defined for each of the sections D1 to D10. As an example, the operation variable A represents an average value of the operation variable α, and the operation variable B represents a standard deviation of the operation variable α.

  FIG. 4 shows a flowchart of the abnormality determination process executed by the management apparatus 10 (FIG. 1) according to the embodiment.

  In step SA1, the management device 10 (FIG. 1) acquires a verification waveform EW that is a time waveform of a plurality of operation variables measured by the work machine 18 (FIG. 1) to be evaluated. The acquired verification waveform EW is stored in the storage device 13 (FIG. 1).

In step SA2, a plurality of waveforms are cut out from the verification waveform EW, and the cut out waveforms are divided into a plurality of sections on the time axis. The verification waveform EW has a shape that approximates the reference waveform shown in FIG. When an abnormality has occurred in the work machine 18, the waveforms of some operating variables show different shapes from the reference waveform. The procedure for extracting a plurality of waveforms from the verification waveform EW is the same as the procedure for extracting a plurality of partial waveforms from the reference waveform. In the present embodiment, similarly to the extraction from the reference waveform, a waveform for a certain period (for example, 10 seconds) including the time when the waveform of the operation variable α rises is extracted from the verification waveform of each of the plurality of operation variables.

  Each of the waveforms cut out from the verification waveform is divided into 10 sections D1 to D10 on the time axis. The number of sections is the same as the number of sections D1 to D10 (FIG. 3) included in one reference data RD. For each of the sections D1 to D10, the statistics of the operating variables, for example, the average value and the standard deviation are obtained. Verification data ED is obtained from each of the cut out waveforms.

  FIG. 5 shows an example of the verification data ED. Each of the verification data ED1, ED2, ED3,... Is composed of a plurality of operation variables A to I defined for each of the sections D1 to D10, like the reference data RD (FIG. 3). In the example shown in FIG. 5, the operation variable A in the section D1 of the verification data ED1 is represented by a (1). The obtained verification data ED1, ED2, ED3,... Are stored in the storage device 13 (FIG. 1).

  In step SA3 of FIG. 4, the discretized feature amount is determined based on the plurality of operation variables A to I for each of the sections D1 to D10 of the reference data RD and the verification data ED. With reference to FIGS. 6-8B, an example of the process of step SA3 is demonstrated.

  FIG. 6 shows a flowchart of step SA3. In step SA301, the standardized reference data RDS is obtained by standardizing the values of the operation variables A to I constituting the reference data RD. The standardized reference data RDS is stored in the storage device 13 (FIG. 1). Hereinafter, a method for obtaining the standardized reference data RDS will be described. First, the average value and standard deviation of each of the operating variables A to I are obtained using all the sections D1 to D10 of all the reference data RD as a population. By converting the values of the operating variables A to I in the sections D1 to D10 of the reference data RD so that the average value of each of the operating variables A to I is 0 and the standard deviation is 1, standardized reference data RDS is obtained.

  In the above example, when standardizing the reference data RD, all the sections D1 to D10 of all the reference data RD are used as a population for obtaining an average value and a standard deviation. It may be excluded from the group. Hereinafter, a method for determining a section to be excluded from the population will be described using the reference data RD shown in FIG. 3 as an example.

  Focusing on the operation variable B of the reference data RD1 shown in FIG. 3, the value of the operation variable B in the sections D5 to D7 is significantly larger than the value of the operation variable B in the other sections D1 to D4 and D8 to D10. When standardization is performed using all values in the sections D1 to D10 as a population, the fluctuations in the operating variable B in the sections D1 to D4 and D8 to D10 are buried in large fluctuations including the sections D5 to D7. When standardization is performed using all values in the sections D1 to D10 as a population, even if there are significant fluctuations in the sections D1 to D4 and D8 to D10, these fluctuations are not substantially reflected in the abnormality determination process. .

By excluding the sections D5 to D7 from the population and standardizing each of the driving variables B, fluctuations in the driving variables B in the sections D1 to D4 and D8 to D10 can be reflected in the abnormality determination process. . In this case, the average values and standard deviations of the operating variables A to I are obtained using the values of the operating variables A to I in the sections D1 to D4 and D8 to 10 of all the reference data RD as a population. Using these average values and standard deviations, the operating variables A to I are normalized. Even when the sections D5 to D7 in which the fluctuation of the driving variable B is larger than the predetermined width are excluded from the population, the values of the driving variables A to I in the sections D5 to D7 are used as the average value and the standard deviation. Use to convert.
Thereby, standardized reference data RDS including values of all the sections D1 to D10 is obtained.

  As described above, for each of the operation variables A to I, a section to be excluded from the population at the time of standardization may be determined based on the distribution of values in the sections D1 to D10. For example, judging from the distribution of values in the sections D1 to D10, sections having unique values may be excluded. Even when a section having a unique value is excluded from the population when calculating the average value and the standard deviation, the standardized reference data RDS includes operating variables of the section having a unique value. For this reason, the unique value of the operating variable is also reflected in the abnormality determination result.

  FIG. 7 shows an example of the standardized reference data RDS obtained by standardizing the operation variables A to I by excluding the sections D5 to D7 from the population. In the reference data RD before standardization shown in FIG. 3, the units of the operating variables A to I are different. For example, the unit of engine torque is “N · m”, the unit of engine speed is “rpm”, and the unit of cooling water temperature is “° C.”. For this reason, no significant information can be obtained even if the values of different operating variables are compared as they are. By standardizing the values of the operation variables A to I, the operation variables A to I become dimensionless, and different operation variables can be compared.

  The verification data ED is standardized using the same average value and standard deviation as the average value and standard deviation used when standardizing the reference data RD. Thereby, standardized verification data EDS is obtained.

  In step SA302 shown in FIG. 6, the standardized reference data RDS and the standardized verification data EDS are input and based on the standardized values of the driving variables A to I for each of the sections D1 to D10 using the clustering method. Then, the discretized feature amount is determined. As a clustering method, for example, a K-average method can be applied.

  FIG. 8A shows an example of the discretized reference data RDD, and FIG. 8B shows an example of the discretized verification data EDD. 8A and 8B show an example in which the number of clusters is ten. For this reason, any one of 10 feature amounts from 0 to 9 is assigned to each of the sections D1 to D10. For example, the feature quantity “2” is assigned to the section D1 of the discretized reference data RDD1. The number of clusters may be 9 or less, or 11 or more. Each of the discretized reference data RDD1, RDD2, RDD3,... And the discretized verification data EDD1, EDD2, EDD3,. For example, the time series of the feature amounts constituting the discretized reference data RDD1 is “22229441222”. The obtained discretized reference data RDD and discretized verification data EDD are stored in the storage device 13.

  In step SA4 of FIG. 4, based on the discretized reference data RDD and the discretized verification data EDD, it is determined whether or not the work machine from which the verification waveform EW is acquired is abnormal. A method of determining whether or not the work machine is abnormal will be described with reference to FIGS. As an example, a hidden Markov model is applied to this determination, and a Baum Welch algorithm is used to search for unknown parameters.

  FIG. 9 shows a flowchart of step SA4. In step SA401, the number of states of the state transition model applied to the hidden Markov model is set. In the embodiment, a left-to-right model in which state transition proceeds in one direction is employed. In addition, the number of states is the simplest two states. The number of states in the state transition model may be 3 or more.

FIG. 10 shows a state transition diagram. The state transitions from the initial state SB to the final state SE via the first state S1 and the second state S2. Since the left-to-right model is employed, the state does not change from the second state S2 to the first state S1 in the reverse direction. Further, the first state S1 is bypassed and the first state S1 is bypassed and no direct transition is made to the second state S2, and the first state S1 is bypassed and the second state S2 is bypassed and no direct transition is made. The self transition probability of the first state S1 is represented by t (1,1), and the transition probability from the first state S1 to the second state S2 is represented by t (1,2). The self transition probability of the second state S2 is represented by t (2, 2), and the transition probability from the second state S2 to the final state SE is represented by t (2, e). In the first state S1, an output probability that an event that outputs the feature quantity i occurs is represented by e (1, i), and in the second state S2, an output probability that an event that outputs the feature quantity i occurs is represented by e (2 , I).

  In step SA402, provisional values of the transition probability t and the appearance probability e of the state transition model are set. An appropriate value may be set as the provisional value.

  In step SA403, a plurality of state sequences K are generated by applying each state of the state transition model to each of the sections D1 to D10 of the discretized reference data RDD.

  FIG. 11 shows a plurality of state series K (1) to K (9) generated for the discretized reference data RDD. Since the number of sections D1 to D10 is 10 and the number of states is 2, the same state is assigned to a plurality of sections. For example, in the state series K (2), the first state S1 is assigned from the section D1 to the section D8, and the second state S2 is assigned to the sections D9 to D10. Since the state transition proceeds only in one direction, that is, in the direction from the first state S1 to the second state S2, the number of generated state sequences K (1) to K (9) is nine.

  In step SA404 in FIG. 9, for each state series K, the generation probability PK of each state series K is calculated using the provisional values of the transition probability t and the appearance probability e of the state transition model. For example, when attention is paid to the state series K (1) shown in FIG. 11, the first state S1 is assigned to the section D1 of the discretized reference data RDD1, and an event that the feature amount is 2 appears. The transition destination state is the first state S1. The probability that an event having a feature amount of 2 appears and makes a transition to the first state S1 is represented by e (1,2) × t (1,1). Here, the transition probability t (1,1) and the appearance probability e (1,2) are defined in FIG.

Similar probabilities can be obtained for the ten states S constituting the state sequence K (1). By multiplying all the probabilities obtained for these ten states S, the generation probability PK (1) of the state series K (1) is calculated. The generation probability PK (1) of the state series K (1) shown in FIG. 11 is expressed by the following equation (1).
PK (i) = e (1,2) × t (1,1)
× e (1,2) × t (1,1)
× e (1,2) × t (1,1)
× e (1,2) × t (1,1)
× e (1,9) × t (1,1)
× e (1,4) × t (1,1)
× e (1,1) × t (1,1)
× e (1,2) × t (1,1)
× e (1,2) × t (1,2)
× e (1,2) × t (2, e) (1)

  Similarly, the generation probabilities PK (2) to PK (9) of the other state series K (2) to K (9) are calculated. Further, the generation probability PK of each state series is calculated for the state series K generated with respect to the other discretized reference data RDD2, RDD3, RDD4,.

In step SA405, the transition probability t and the appearance probability e are calculated based on the generation probability PK of the state series K generated for all the discretized reference data RDD, and the provisional values are replaced with newly calculated values. . Specifically, the feature quantity is 0 for all the first states S1 appearing in all the state sequences K assigned to all the discretized reference data RDD1, RDD2, RDD3,... Shown in FIG. The number of appearances of events taking ˜9 is weighted by the generation probability PK and totaled. Based on the total number of times, each of the appearance probabilities e (1, 0) to e (1, 9) can be calculated. By executing the same procedure for all the second states S2, the appearance probabilities e (2, 0) to e (2, 9) can be calculated.

  Further, for all first states S1 appearing in all state sequences K assigned to all the discretized reference data RDD1, RDD2, RDD3,... Shown in FIG. The number of transitions to the two-state S2 is weighted with the generation probability PK and totaled. Based on the total number of times, the transition probabilities t (1,1) and t (1,2) can be calculated. Similarly, transition probabilities t (2, 2) and t (2, e) can be calculated.

  In step SA406, it is determined whether or not the generation probabilities PK of the state series K have converged. Specifically, when the difference between the value of the generation probability PK before calculating the generation probability PK in step SA404 and the value of the generation probability PK after calculation is smaller than the convergence determination value, the generation probability PK has converged. It is determined. If it is determined that the generation probability PK has not converged, the process returns to step SA404, and the generation probability PK is calculated using the new provisional value.

  If it is determined that the generation probability PK has converged, the nearest provisional values of the transition probability t and the appearance probability e are adopted as definite values in Step SA407. This determined value is stored in the storage device 13 (FIG. 1) as a part of the parameters that define the state transition model.

  In step SA408, the generation probability PK of the state sequence K generated corresponding to each of the discretization verification data EDD1, EDD2, EDD3,... Calculate based on Hereinafter, a method for calculating the generation probability PK will be described.

  As shown in FIG. 12, a plurality of state series K is generated by assigning the first state S1 or the second state S2 of the state transition model to the sections D1 to D10 of the discretization verification data EDD. The method for generating the state series K is the same as the method for generating the state series K for the discretized reference data RDD shown in FIG. The generation probability PK of each state series K is calculated by applying the determined values of the transition probability t and the appearance probability e to the state series K. For calculating the generation probability PK, an expression similar to the above expression (1) is used.

In step SA409, for each of the plurality of discretized reference data RDD and the plurality of discretized verification data EDD, the generation probabilities PK (1) to (9) of the state series K (1) to (9) are added together. The total generation probability PT is obtained. The score SC is obtained by inverting the sign of the natural logarithm of the total generation probability PT. The score SC is defined by the following formula.
SC = -ln (PT)
A common logarithm may be used instead of the natural logarithm.

  In step SA410, it is determined based on the score SC obtained for each of the reference data RD and the verification data ED whether or not the work machine to be evaluated is abnormal. Hereinafter, a method of determining whether or not the work machine is abnormal will be described with reference to FIG.

FIG. 13 shows the score SC obtained for each of the reference data RD and the verification data ED. The horizontal axis corresponds to the reference data RD and the verification data ED, and the vertical axis represents the score SC.
In the example illustrated in FIG. 13, the scores SC obtained for the reference data RD are distributed between 15 and 25, and the scores SC obtained for the verification data ED are distributed between 30 and 36. In this example, it can be said that there is a significant difference between the average values of the two. As described above, when there is a significant difference between the average value of the score SC obtained for the verification data ED and the average value of the score SC obtained for the reference data RD, the work machine to be evaluated It is determined that some abnormality has occurred. If there is no significant difference between the two, it is determined that the work machine to be evaluated is normal.

  Next, an example of a specific determination method for determining whether there is a significant difference will be described. An average value and a standard deviation of the score SC obtained for the reference data RD are obtained. The difference between the average value of the score SC obtained for the reference data RD and the average value of the score SC obtained for the verification data ED is compared with the standard deviation of the score obtained for the reference data RD. When the difference between the average values is equal to or greater than the criterion value set based on the standard deviation, the average value of the scores SC obtained for the reference data RD and the average value of the scores SC obtained for the verification data ED Is determined to have a significant difference. For example, n times the standard deviation is adopted as the determination reference value. Here, n is a positive real number, for example, n = 2.

  The work machine management device 10 according to the above embodiment determines whether or not an abnormality has occurred in the work machine based on the feature values of the operation variables arranged in time series. If the operating variable is integrated with respect to time, the information regarding the short time fluctuation of the operating variable is lost. On the other hand, in the embodiment, since the abnormality determination of the work machine is performed based on the feature amount of the operation variable arranged in time series, whether or not the work machine is abnormal in consideration of the abnormality of the time variation of the operation variable. Can be determined. Thereby, abnormality detection accuracy can be improved.

  In the above embodiment, the reference data RD (FIG. 4) is the data before being discretized as shown in FIG. As the reference data RD, the discretized data shown in FIG. 8A may be stored in the storage device 13 (FIG. 1). In this case, in step SA3 (FIG. 4), the verification data ED may be discretized using the same clustering method as that applied to discretize the reference data RD.

  Although the present invention has been described with reference to the embodiments, the present invention is not limited thereto. It will be apparent to those skilled in the art that various modifications, improvements, combinations, and the like can be made.

DESCRIPTION OF SYMBOLS 10 Management apparatus 11 Input / output apparatus 12 Processing apparatus 13 Storage apparatus 18 Work machine 19 Communication line 20 Sensor 21 Control apparatus 22 Communication apparatus D1-D10 Section RD Reference data RDS Standardization reference data RDD Discretization reference data EW Verification waveform ED Verification data EDS Normalization verification data EDD Discretization verification data K State sequence PK State sequence generation probability SC Score t Transition probability e Output probability

Claims (10)

A storage device for storing a plurality of reference data representing time waveforms at the time of normal operation of a plurality of operation variables measured by sensors mounted on the work machine and depending on an operating state of the work machine;
A processing device,
The processor is
The time waveform of the operation variable obtained by measuring the plurality of operation variables of the work machine to be evaluated is divided into a plurality of sections on the time axis, and the time of the plurality of operation variables for each section. Obtain verification data including values calculated based on the waveform,
For each of the sections, by determining the discretized feature value based on the value of the operation variable of the verification data, the discretization verification data including the time series of the feature value is obtained,
A work machine management device that determines whether or not the work machine to be evaluated is abnormal based on the reference data and the discretization verification data. The processor is
The reference data is divided into a plurality of sections on the time axis, and for each section, the value of the driving variable is calculated based on the time waveforms of the plurality of driving variables,
For each of the sections, a plurality of discretized reference data consisting of a time series of the discretized feature values is obtained based on the reference data,
The work machine management device according to claim 1, wherein a determination is made as to whether or not the work machine to be evaluated is abnormal based on the discretized reference data and the discretized verification data.   3. The work machine management device according to claim 2, wherein the processing device obtains a plurality of the discretized reference data based on values of the operation variables in a remaining section obtained by removing some sections from the plurality of sections. . The processor is
The value of the operation variable included in the reference data and the value of the operation variable included in the verification data are normalized for each operation variable based on an average value and a standard deviation,
By using a standardized value of the driving variable of the reference data and a standardized value of the driving variable of the verification data as an input, using the clustering technique, the discretization verification data and The work machine management device according to claim 2, wherein the discretized reference data is obtained. The processor is
When standardizing the value of the operation variable included in the reference data, the value of the operation variable is determined for each operation variable, excluding the value of the operation variable in some of the sections from the plurality of sections. Find the mean and standard deviation,
The work machine management device according to claim 4, wherein the operation variable value of the reference data is standardized based on the obtained average value and standard deviation. The processor is
Applying a plurality of the discretized reference data to a state transition model including a transition probability between states and an appearance probability of an event having a characteristic value for each state, the transition probability and the appearance probability are obtained. And determining the generation probability of each of the discretized reference data,
Based on the state transition model, the transition probability, and the appearance probability, the generation probability of the discretization verification data is obtained,
The determination according to any one of claims 2 to 5, wherein whether or not the work machine to be evaluated is abnormal is determined based on a generation probability of each of the discretized reference data and a generation probability of the discretized verification data. The work machine management device described.   The processing device determines whether or not the work machine to be evaluated is abnormal based on a logarithm of the generation probability of each of the discretized reference data and a logarithm of the generation probability of the discretized verification data. The work machine management device according to 6. The processor is
(A) setting a provisional value as the transition probability and the appearance probability of the state transition model;
(B) generating a plurality of state series by applying each state of the state transition model to the feature quantity constituting each of the discretized reference data;
(C) For each state series, using the provisional value as the transition probability and the appearance probability of the state transition model, calculate the generation probability of the state series,
(D) resetting the provisional value set as the transition probability and the appearance probability of the state transition model based on the calculated generation probability of the state series;
(E) The transition probability and the appearance probability of the state transition model are calculated by repeating the processes (c) and (d) until the generation probability of the state series converges. The work machine management device described. As the reference data,
A time waveform of the operation variable obtained by measurement during normal operation of the work machine is divided into a plurality of sections on the time axis, and is discrete based on the values of the plurality of operation variables for each section. The work machine management device according to claim 1, wherein discretized reference data configured by the time series of the converted feature quantities is stored in the storage device. The time waveform of the operating variable , which is obtained by measurement with the sensor mounted on the work machine to be evaluated and depends on the operating state of the work machine, is divided into multiple sections on the time axis, and is obtained for each section. Obtaining verification data including values of a plurality of the operating variables;
Determining discretized verification data consisting of a time series of the feature values by determining a discretized feature value for each section based on the value of the operation variable of the verification data;
Whether the work machine to be evaluated is abnormal based on a plurality of reference data representing a time waveform of the operating variable at a normal time obtained during normal operation of the work machine and the discretized verification data. An abnormality determination method for a work machine having a step of determining whether or not.

Images