JP2020160870A - Troubleshooting system having notification feature - Google Patents

Abstract

To immediately notify a serviceperson of an event occurring in a work machine, and explicitly instruct the serviceperson on a treatment that copes with the event.SOLUTION: A troubleshooting system S includes a forklift 1, a server 2, and a terminal 3. The forklift 1 transmits to the server 2 output data of a sensor, which detects a moving condition of the forklift 1, and internal data of control software that controls the forklift 1. The server 2 detects an even occurring in the forklift 1 on the basis of the received output data of the sensor and the received internal data of the control software, and transmits to the terminal 3, which a serviceperson P1 uses, a list of low-level events associated with the detected event or a treatment together with the event occurring in the forklift 1.SELECTED DRAWING: Figure 1

Description

Regarding troubleshooting systems with notification functions for work machines.

When a serviceman related to a work machine such as a crane car or a forklift has a trouble such as an accident or a breakdown in the work machine, he / she takes some measures to solve the event of the work machine caused by the trouble. Skilled service personnel are familiar with the treatment of various events in the work machine. On the other hand, inexperienced servicemen often do not know the appropriate treatment for this event, and it is difficult to easily recognize the appropriate treatment.

Therefore, for example, there is a failure diagnosis method as disclosed in Patent Document 1. The failure diagnosis unit of the machine tool control device disclosed in Patent Document 1 includes a unit storage unit, a phenomenon storage unit, a defect occurrence prediction component storage unit, a treatment method storage unit, a solution storage unit, and a failure diagnosis control. It has a department. The operator selects the unit in which the defect has occurred using the display unit consisting of the touch panel, selects the failure phenomenon corresponding to each unit, selects the component that can cause the defect corresponding to each phenomenon, and selects each component. Select the corresponding action method and narrow down the solution to the problem. As a result, even an inexperienced worker can respond to an event of a machine tool.

By the way, when the operator recognizes that a trouble has occurred in the work machine, there is a problem that a serious event has already occurred in the work machine. Moreover, in the case of an unmanned work machine that works unmanned, even if a trouble occurs, the administrator may not be nearby. In this case, since the administrator later recognizes the trouble and then contacts the serviceman, there is also a problem that the situation worsens with the passage of time. Therefore, when a trouble occurs in the work machine, it is preferable to promptly notify the service person or other related persons of the work machine of the event.

JP-A-2017-68678

In view of the above circumstances, the problem to be solved by the present invention is a troubleshooting system capable of promptly notifying a serviceman of an event that has occurred in a work machine and clearly indicating to the serviceman a measure corresponding to this event. Is to provide.

In order to solve the above problems, the troubleshooting system provided with the notification function for the work machine according to the present invention is
With the work machine
A storage unit that stores events that can occur in the work machine and actions corresponding to the events, and
A display unit that displays the entered information and
An input unit for inputting event information consisting of an event occurring in the work machine or a part of the event,
An output unit that acquires a list or action of lower-level events corresponding to the input event information from the storage unit and outputs the list to the display unit.
Event detector and
With a notification section,
The work machine has a data transmission unit that transmits output data of a sensor that detects the operating state of the work machine and internal data of control software that controls the work machine to the event detection unit.
The event detection unit detects an event that has occurred in the work machine based on the received output data of the sensor and the internal data of the control software.
The notification unit acquires a list or treatment of lower-level events corresponding to the event detected by the event detection unit from the storage unit, and the serviceman or the work together with the event generated in the work machine. It is characterized by notifying other parties to the machine and / or both.

The above troubleshooting system is preferably
The event detection unit further detects an event having a probability of occurrence equal to or higher than a predetermined value based on the received output data of the sensor and the internal data of the control software.
The notification unit further acquires a list or treatment of lower-level events corresponding to the event having an occurrence probability of a predetermined value or more detected by the event detection unit from the storage unit, and before the occurrence probability is a predetermined value or more. Notify the serviceman and / or other parties of the work machine along with the event.

The above troubleshooting system is preferably
When the correlation is learned by using the combination of the output data of the sensor, the internal data of the control software, and the event generated in the work machine as teacher data, and the output data of the sensor and the internal data of the control software are input. Further equipped with a trained first neural network that outputs an event that has occurred in the work machine or has an occurrence probability of a predetermined value or more, or both.
The event detection unit uses the trained first neural network to detect an event that has occurred in the work machine from the output data of the sensor and the internal data of the control software, or an event having a probability of occurrence of a predetermined value or more, or both. To detect.

The above troubleshooting system is, for example,
The notification unit further notifies the event having a probability of occurrence of a predetermined value or more as important information, and notifies the event that has occurred in the work machine as emergency information.

The above troubleshooting system is preferably
Further comprising a terminal and a server capable of communicating with the terminal and the work machine.
The terminal has the display unit, the input unit, and the output unit.
The server has the storage unit, the event detection unit, and the notification unit.
The terminal is used by the serviceman and
The notification unit performs the notification to the service person by transmitting the notification to the terminal.
The display unit further displays a list or action of an event received by the terminal and a lower-level event corresponding to the event.

The above troubleshooting system is preferably
A similarity calculation unit for calculating the similarity between the input event information and the event stored in the storage unit is further provided.
The output unit further outputs a list of the events for which a degree of similarity equal to or higher than a predetermined value has been calculated to the display unit.

The above troubleshooting system is preferably
Unsupervised learning is performed using a neural network using a plurality of past events that have occurred in the work machine described in a plurality of reports as learning data, and when a word is input, it corresponds to the meaning of the input word. A trained second neural network that outputs a distributed representation vector to
When the event information is input, the morphological analysis unit that divides the event information into words and
A word vectorization unit that converts event information divided for each word into a distributed expression vector for each word by the learned second neural network.
An event vectorization unit that converts the input event information into a distributed expression vector in units of event information based on the distributed expression vector for each word included in the event information.
A vector expression storage unit that stores the event in the storage unit as a distributed expression vector is further provided.
The similarity calculation unit determines the cosine similarity between the event information vectorized by the event vectorization unit and the distributed expression vectorized event stored in the vector expression storage unit. Based on this, the similarity is calculated.

The above troubleshooting system is preferably
The distributed representation vector corresponding to the meaning of the input word is based on the distributed representation vector of the subword extracted by dividing the input word into N-grams.

The troubleshooting system according to the present invention can promptly notify the serviceman of an event that has occurred in the work machine, and can clearly indicate to the serviceman the action corresponding to this event.

It is an overall view which shows the schematic structure of one Embodiment of the troubleshooting system which concerns on this invention. It is a block diagram of the troubleshooting system of FIG. It is a figure which shows an example of the report which described the past event. It is a figure which shows an example of the input screen displayed on the display part of FIG. It is a figure which shows another example of an input screen. This is an example of the treatment displayed on the display unit of FIG.

Hereinafter, an embodiment of a troubleshooting system having a notification function according to the present invention will be described with reference to the drawings. The work machine in this embodiment is a forklift 1, but it is merely an example and is not limited thereto.

FIG. 1 is an overall view showing a schematic configuration of the troubleshooting system S according to the present embodiment. The troubleshooting system S has its operating state transmitted from the forklift 1 to the server 2 at any time via the network 100, and the server 2 analyzes the operating state of the forklift 1 to detect an event occurring in the forklift 1 and promptly detect it. It is configured to notify the serviceman P1. Hereinafter, each configuration will be specifically described.

<Forklift>
As shown in FIG. 2, the forklift 1 includes a sensor 10, control software 11, and a data transmission unit 12.

The sensor 10 is for detecting the operation of the forklift 1, and is provided in each unit included in the forklift 1. The sensor 10 may include, for example, any or all of a torque sensor, a vibration sensor, an imaging sensor, a distance sensor, a position sensor, a temperature sensor, a humidity sensor, a flow rate sensor, a light amount sensor, and a pressure sensor. Further, the sensor 10 may include, for example, an ambient environment sensor that detects the ambient environment state of the forklift 1.

The control software 11 is software for controlling the forklift 1. For example, the control software 11 controls the operation of the cylinder, the motor, the steering unit, and the like. The internal data of the control software 11 may include, for example, any or all of torque, position, velocity, acceleration, current and voltage. Further, the internal data of the control software 11 includes control command information. The control command information is the content of commands to each unit by the control software 11, and corresponds to, for example, a forward rotation command to a motor, a stop command, and the like.

The data transmission unit 12 transmits the output data of the sensor 10 and the internal data of the control software 11 to the server 2 at any time via the network 100. The data transmission unit 12 may accumulate data for a certain period of time and transmit the accumulated data at any time. As a result, for example, the transition of the moving speed of the forklift 1 after the stop command to the motor by the control software 11 is transmitted.

<Server>
The server 2 can communicate with the terminal 3 and the forklift 1 via the network 100. The server 2 includes a storage unit 20, a learned first neural network (hereinafter referred to as “learned first NN”) 21, an event detection unit 22, a notification unit 23, a morphological analysis unit 24, and a word vectorization unit. It has 25, a learned second neural network (hereinafter, referred to as “learned second NN”) 26, an event vectorization unit 27, a similarity calculation unit 28, and a vector expression storage unit 29.

The storage unit 20 stores an event that may occur in the forklift 1 and a measure corresponding to this event. The event and the action are extracted from a plurality of trouble cases including the event that occurred in the forklift 1 in the past and the action taken for the event. Events and actions may be extracted, for example, from trouble cases in Report 300 (see FIG. 3) described by Serviceman P1. In addition, the storage unit 20 also stores the causal relationship between the events, whereby the upper event and the lower event are associated with each other.

When the output data of the sensor 10 and the internal data of the control software 11 are input, the trained first NN 21 outputs an event that has occurred in the forklift 1 or has a probability of occurrence of a predetermined value or more, or both. The trained first NN21 learns the correlation in advance using the combination of the output data of the sensor 10, the internal data of the control software 11 and the event generated in the forklift 1 as the teacher data, and the output data of the sensor 10 and the control software 11 When the internal data is input, an event that has occurred in the forklift 1 or has a probability of occurrence or more is output. The trained first NN21 may utilize deep learning and may utilize a convolutional neural network, a recursive neural network, or both. Further, the trained first NN21 may be an ensemble-learned one composed of a plurality of neural networks.

The event detection unit 22 uses the learned first NN 21 to detect an event that has occurred in the forklift 1 from the output data of the sensor 10 and the internal data of the control software 11, or an event having a probability of occurrence of a predetermined value or more, or both. An event having a predetermined occurrence probability or more includes an event having a predetermined probability of occurring within a predetermined time or a predetermined number of days, an event having an already occurring probability of a predetermined value or more, or both.

The notification unit 23 acquires a list of lower-level events or actions corresponding to this event together with the event detected by the event detection unit 22 from the storage unit 20, and notifies the serviceman P1 and other related parties P2 of the forklift 1. .. As a result, the serviceman P1 and other related parties P2 can quickly recognize the event that has occurred or is likely to occur in the forklift 1. In addition, since this notification includes a list of subordinate events or actions corresponding to this event together with the event detected by the event detection unit 22, even an inexperienced service person can quickly respond to the notified event. can do.

The other related party P2 corresponds to, for example, the manager or operator of the forklift 1. The notification unit 23 may notify the serviceman P1 and other related parties P2 of only the events detected by the event detection unit 22. Further, the notification unit 23 distributes the notification destination based on the event detected by the event detection unit 22, whether to notify both the serviceman P1 and other related parties P2 or only the serviceman P1. You may.

The notification method by the notification unit 23 notifies the serviceman P1 and the other related parties P2 by transmitting the notification to the terminal 3 used by the serviceman P1 and the computer used by the other related parties P2. The notification method by the notification unit 23 may be, for example, an e-mail, and is not particularly limited. The notification unit 23 notifies as emergency information if the event detected by the event detection unit 22 is an event that occurred in the forklift 1, and notifies as important information if the event that the probability of occurring in the forklift 1 is equal to or higher than a predetermined value. May be good.

When the event information is input by the input unit 31 (see the terminal 3 in FIG. 2), the morphological analysis unit 24 divides the input event information into words. The event information is a sentence such as "the engine does not start" or a group of words or words expressed as keywords such as "engine" and "does not start" for the event that occurred in the forklift 1. The event information is divided by the morphological analysis unit 24 with a space for each word, and is divided and processed. For example, the event information that "the engine does not start" is divided into "engine", "ga", "starting", and "not". For example, "Mecab" may be used as a tool for morphological analysis. Further, a pre-made corpus for the forklift 1 may be used as a morphological analysis tool. This is preferable because each noun in the event information is appropriately extracted.

The word vectorization unit 25 converts the event information divided for each word by the morphological analysis unit 24 into a distributed expression vector for each word by the learned second NN 26.

When a word is input, the learned second NN26 outputs a distributed expression vector corresponding to the meaning of the input word. The learned second NN26 is unsupervised-learned in advance using an event described in a plurality of past reports 300 divided for each word as learning data using a neural network, and the event divided for each word. When the information is input, the distributed expression vector corresponding to the meaning of each word is output based on the relationship between each word included in the event information.

In the learned second NN26, when words with high relevance such as words that are likely to appear in the same event (co-occurrence) or words that are likely to appear in close positions in the same event are input. It has been pre-learned to output a similar distributed representation vector. This is based on the distribution hypothesis (Harrs, 1954, Rubenstein & Goodenough, 1965) that words appearing in similar contexts are also semantically similar. As the word vectorization unit 25, for example, a language model such as “Word2Vec”, “Glove”, or “Fastext” based on neural network technology may be used. For distributed representations and Word2Vec, see "MIKOLOV, Tomas, et al. Distributed representations of words and phrases and their compositionality. In: Advances in neural information processing systems. 2013. p. 3111-3119." For Grove, see "PENNINGTON, Jeffrey; SOCHER, Richard; MANNING, Christopher D. Glove: Global Vectors for Word Representation. In: EMNLP. 2014. p. 1532-1543." For Fastext, you can refer to "JOULIN, Armand, et al. Bag of tricks for efficient text classification. ArXiv preprint arXiv: 1607.01759, 2016."

As a method of outputting the distributed expression vector of each word, the learned second NN26 preferably creates a distributed expression vector in consideration of the subwords included in each word. Specifically, the word vectorization unit 25 divides the input word into N-grams, extracts subwords, and uses the distributed expression vector corresponding to the meaning of the input word as the sum of the distributed expression vectors of the subwords or It is preferable to use an average. For example, in the case of the word "intake hose", the subword corresponds to "suction", "intake", "kiho", "ho", "su", and "su". As a result, the learned second NN26 outputs similar distributed expression vectors to, for example, the “intake hose”, the “suction hose”, and the “hose”, so that the input event information is swayed and unknown words. Can also be supported. For example, "Fastext" may be used as a tool for creating a distributed representation vector in consideration of subwords.

As shown in FIG. 3, in the report 300, the event that occurred in the forklift 1 is described in the defect content column by each person in charge (serviceman P1). Therefore, in the report 300, each event is described in various expressions in combination with the individuality of the person. By using the description of these events as learning data, the learned second NN26 can appropriately convert each word included in the event information of various expressions into a distributed expression vector.

The event vectorization unit 27 is used by the word vectorization unit 25 for each word so that the similarity calculation unit 28 can calculate the similarity between the input event information and the event stored in the storage unit 20. Distributed representation vectorized event information is converted into a distributed representation vector of event information units based on the distributed representation vector for each word. The event vectorization unit 27 may use the average of the distributed expression vectors of the words included in the event information as the feature vector of the event information itself as a method for converting the event information into a distributed expression vector. Alternatively, the event vectorization unit 27 may use the feature vector averaging using SCDV (Sparse Composite Document Vectors) as a method for more accurately converting the event information into a distributed representation vector. For SCDV, see "Dheeraj Mekala et al .: SCDV: Sparse Composite Document Vectors using soft clustering over distributional representations, Proc of EMNLP, 2017".

The similarity calculation unit 28 calculates the similarity between the input event information and the event stored in the storage unit 20. Events in the storage unit 20 are morphologically analyzed in advance by the morphological analysis unit 24, decentralized expression vectorization for each word by the word vectorization unit 25, and decentralized expression vectorization for each event by the event vectorization unit 27. There is.

The vector expression storage unit 29 stores the events in the storage unit 20 that are distributed and expressed vectorized by the event vectorization unit 27 in association with each event in the storage unit 20.

The similarity calculation unit 28 receives input event information based on the cosine similarity between the distributed expression vectorized event information and the distributed expression vectorized event stored in the vector expression storage unit 29. The degree of similarity with the event in the storage unit 20 is calculated. Cosine similarity is a similarity calculation method used when comparing documents in a vector space model. Since the cosine similarity expresses the closeness of the angles formed by the vectors, the input event information and the event in the storage unit 20 are more similar if the cosine similarity is close to 1, and should be close to 0. It will be less similar.

<Terminal>
The terminal 3 has a display unit 30, an input unit 31, and an output unit 32. Although the terminal 3 is a tablet in the present embodiment, it is merely an example, and may be, for example, a laptop computer or a smartphone.

The display unit 30 displays the input information. The display unit 30 corresponds to a tablet display. In addition, the display unit 30 displays a list or action of the event received from the notification unit 23 and the lower-level event corresponding to this event.

The input unit 31 is a tool for inputting event information and the like. In the case of a tablet, the display is a touch panel, which also serves as a display unit 30 and an input unit 31. The input unit 31 may be, for example, a mouse or a keyboard. Basic information and event information including any or all of serviceman information, date and time information, place information, and customer information regarding the serviceman P1 are input by the input unit 31. Further, an event (event information) is selected and input by the input unit 31 from the event list displayed on the display unit 30 by the output unit 32. Further, an event (event information) is selectively input by the input unit 31 from the event list transmitted from the server 2 and displayed on the display unit 30.

The output unit 32 acquires a list of lower-level events or actions corresponding to the input event information from the storage unit 20 and outputs the list to the display unit 30. In addition, the output unit 32 also outputs to the display unit 30 a list of events for which a degree of similarity of a predetermined value or higher is calculated by the degree of similarity calculation unit 28. Further, when an event is selected and input from the event list, the output unit 32 acquires a list of events or actions corresponding to the input event from the storage unit 20 and outputs the list to the display unit 30.

Next, troubleshooting by the terminal 3 will be specifically described with reference to the drawings.

From the serviceman P1, basic information and the like are input to the terminal 3 by the input unit 31.

As shown in FIG. 4, when the screen 400 for inputting the forklift 1 information and the event information is displayed on the display unit 30, the serviceman P1 sets the brand, type, and model of the forklift 1 in the image areas 400a to 400d. The model, the part where the event occurred, and the error code of the event are directly input or selectively input.

Next, as shown in FIG. 5, a screen 500 for selecting a top event is displayed on the display unit 30, and when the event information 500a and 500b are displayed, one of the event information is selected by the serviceman P1. Entered. At this time, the event information corresponding to the event that occurred in the forklift 1 may not be included in the displayed event information. In this case, the serviceman P1 inputs the event information to the keyword input area 400e shown in FIG.

The similarity calculation unit 28 calculates the similarity between the input event information and the event stored in the storage unit 20, and transmits the event for which the similarity equal to or higher than a predetermined degree is calculated to the terminal 3. The output unit 32 outputs a list of received events to the display unit 30 and displays it. Next, when the serviceman P1 selects and inputs an event corresponding to the event occurring in the forklift 1 from the displayed event list, the output unit 32 stores a list of events corresponding to the input event in the storage unit 20. Is received from and output to the display unit 30.

As shown in FIG. 6, when the serviceman P1 selectively inputs the lowest event, the output unit 32 outputs and displays the treatment 600a corresponding to this event on the display unit 30. Even an immature serviceman P1 can identify a treatment corresponding to an event occurring in the forklift 1 by referring to this treatment.

<A series of troubleshooting system operations>
The configurations of the troubleshooting system S according to the present embodiment have been described above. The series of operations of the forklift 1, the server 2, and the terminal 3 in the troubleshooting system S can be summarized as follows.

(1) The forklift 1 transmits its operating state to the server 2 at any time by the data transmission unit 12.

(2) When the event detection unit 22 detects an event that has occurred in the forklift 1 or an event whose probability of occurrence is equal to or higher than a predetermined value, the server 2 notifies the serviceman P1 and other related parties P2 by the notification unit 23.

(3) When the terminal 3 receives an event that has occurred in the forklift 1 from the server 2, the terminal 3 displays the received event on the display unit 30, and displays the action corresponding to this event on the display unit 30. As a result, even an inexperienced serviceman P1 can immediately respond to the event of forklift.

(4) When the terminal 3 receives an event from the server 2 that has a probability of occurring in the forklift 1 or more, the terminal 3 displays the received event on the display unit 30, and displays the action corresponding to this event on the display unit 30. indicate. As a result, the serviceman P1 can prevent the occurrence of troubles in the forklift 1 by taking measures in advance before the event occurs.

(5) When the terminal 3 receives the event list together with the detected event, the terminal 3 displays the lower event list of this event together with the detected event on the display unit 30.

(6) Next, in the terminal 3, the serviceman P1 specifies which event in the lower-level event list is occurring in the forklift 1, and the specified event (event information) is input. ..

(7) Next, the terminal 3 acquires the action corresponding to the event specified by the serviceman P1 from the storage unit 20 of the server 2 and displays it on the display unit 30. As a result, the serviceman P1 can perform the displayed procedure on the forklift 1.

Although one embodiment of the troubleshooting system S according to the present invention has been described above, the present invention is not limited to the above embodiment.

(1) In the troubleshooting system S, the terminal 3 may have the configuration of the server 2. In this case, the transmission unit transmits the output data of the sensor 10 and the internal data of the control software 11 to the terminal 3.

(2) The storage unit 20 may further store technical data corresponding to the treatment. In this case, the output unit 32 may output the technical data corresponding to this treatment to the display unit 30 together with the treatment and display it on the display unit 30. As a result, even an inexperienced serviceman P1 can take appropriate measures.

(3) The event detection unit 22 may use another trained model as long as it can detect an event that has occurred in the forklift 1 using the data received from the data transmission unit.

(4) The event detection unit 22 may be provided on the forklift 1. In this case, the data transmission unit 12 transmits the output data of the sensor 10 and the internal data of the control software 11 to the event detection unit 22. Further, the event detected by the event detection unit 22 is transmitted to the server 2 via the network 100, and the notification unit 23 provides a list or action of this event and a lower-level event corresponding to this event to the serviceman P1 or the forklift 1. Other parties P2 or both are notified.

1 Forklift 10 Sensor 11 Control software 12 Data transmission unit 2 Server 20 Storage unit 21 Learned first neural network 22 Event detection unit 23 Notification unit 24 Morphological analysis unit 25 Word vectorization unit 26 Learned second neural network 27 Event vectorization Unit 28 Similarity calculation unit 29 Vector expression storage unit 3 Terminal 30 Display unit 31 Input unit 32 Output unit

In order to solve the above problems, the troubleshooting system provided with the notification function for the work machine according to the present invention is
With the work machine
A storage unit that stores events that can occur in the work machine and actions corresponding to the events, and
A display unit that displays the entered information and
An input unit for inputting event information consisting of an event occurring in the work machine or a part of the event,
When the event information is input, the morphological analysis unit that divides the event information into words and
Multiple past events that occurred in the work machine described in a plurality of reports are used as learning data, and the relationship between each word in the report is unsupervised learning using a neural network and divided into words. When the event information is input, it corresponds to the meaning of the input word based on the relationship between the learned words and the relationship between the words included in the event information. A trained vector output neural network that outputs a distributed representation vector to
A word vectorization unit that converts event information divided for each word by the morphological analysis unit into a distributed expression vector for each word by the learned vector output neural network.
An event vectorization unit that converts the input event information into a distributed expression vector in units of event information based on the distributed expression vector for each word included in the event information.
The event in the storage unit is previously subjected to morphological analysis by the morphological analysis unit, distributed expression vectorization for each word by the word vectorization unit, and distributed expression vectorization for each event by the event vectorization unit. Is stored in association with each event in the storage unit, and the vector representation storage unit,
The input is based on the cosine similarity between the event information vectorized by the event vectorization unit and the distributed expression vectorized event stored in the vector expression storage unit. A similarity calculation unit that calculates the similarity between the event information and the event stored in the storage unit,
Before the list of lower-level events or actions corresponding to the input event information is acquired from the storage unit, output to the display unit , and the similarity degree above a predetermined value is calculated by the similarity calculation unit. An output unit that outputs a list of the events described to the display unit ,
Event detector and
With a notification section,
The work machine has a data transmission unit that transmits output data of a sensor that detects the operating state of the work machine and internal data of control software that controls the work machine to the event detection unit.
The event detection unit detects an event that has occurred in the work machine based on the received output data of the sensor and the internal data of the control software.
The notification unit acquires a list of lower-level events or measures corresponding to the events detected by the event detection unit from the storage unit, and the serviceman or the work together with the events generated in the work machine. It is characterized by notifying other parties to the machine and / or both.

<Server>
The server 2 can communicate with the terminal 3 and the forklift 1 via the network 100. The server 2 includes a storage unit 20, a learned first neural network (hereinafter referred to as “learned first NN”) 21, an event detection unit 22, a notification unit 23, a morphological analysis unit 24, and a word vectorization unit. It has 25, a learned second neural network (hereinafter, referred to as “learned second NN”) 26, an event vectorization unit 27, a similarity calculation unit 28, and a vector expression storage unit 29. The trained second NN corresponds to the "trained vector output neural network" of the present invention.

Claims (8)

A troubleshooting system with a notification function for work machines
With the work machine
A storage unit that stores events that can occur in the work machine and actions corresponding to the events, and
A display unit that displays the entered information and
An input unit for inputting event information consisting of an event occurring in the work machine or a part of the event,
An output unit that acquires a list or action of lower-level events corresponding to the input event information from the storage unit and outputs the list to the display unit.
Event detector and
With a notification section,
The work machine has a data transmission unit that transmits output data of a sensor that detects the operating state of the work machine and internal data of control software that controls the work machine to the event detection unit.
The event detection unit detects an event that has occurred in the work machine based on the received output data of the sensor and the internal data of the control software.
The notification unit acquires a list of lower-level events or measures corresponding to the events detected by the event detection unit from the storage unit, and the serviceman or the work together with the events generated in the work machine. A troubleshooting system characterized by notifying other parties to the machine and / or both. The event detection unit further detects an event having a probability of occurrence equal to or higher than a predetermined value based on the received output data of the sensor and the internal data of the control software.
The notification unit further acquires a list or treatment of lower-level events corresponding to the event having an occurrence probability of a predetermined value or more detected by the event detection unit from the storage unit, and before the occurrence probability is a predetermined value or more. The troubleshooting system according to claim 1, wherein the service person and / or other related parties of the work machine are notified together with the event. When the correlation is learned by using the combination of the output data of the sensor, the internal data of the control software, and the event generated in the work machine as teacher data, and the output data of the sensor and the internal data of the control software are input. Further equipped with a trained first neural network that outputs an event that has occurred in the work machine or has an occurrence probability of a predetermined value or more, or both.
The event detection unit uses the trained first neural network to detect an event that has occurred in the work machine from the output data of the sensor and the internal data of the control software, or an event having a probability of occurrence of a predetermined value or more, or both. The troubleshooting system according to claim 2, characterized in that it detects. The second or third aspect of the present invention, wherein the notification unit further notifies the event having a probability of occurrence of a predetermined value or more as important information, and notifies the event that has occurred to the work machine as emergency information. Troubleshooting system. Further comprising a terminal and a server capable of communicating with the terminal and the work machine.
The terminal has the display unit, the input unit, and the output unit.
The server has the storage unit, the event detection unit, and the notification unit.
The terminal is used by the serviceman and
The notification unit performs the notification to the service person by transmitting the notification to the terminal.
The troubleshooting system according to any one of claims 1 to 4, wherein the display unit further displays a list of events received by the terminal and a list of lower-level events corresponding to the events or actions. .. A similarity calculation unit for calculating the similarity between the input event information and the event stored in the storage unit is further provided.
The troubleshooting system according to any one of claims 1 to 5, wherein the output unit further outputs a list of the events for which a degree of similarity of a predetermined value or more has been calculated to the display unit. Unsupervised learning is performed using a neural network using a plurality of past events that have occurred in the work machine described in a plurality of reports as learning data, and when a word is input, it corresponds to the meaning of the input word. A trained second neural network that outputs a distributed representation vector to
When the event information is input, the morphological analysis unit that divides the event information into words and
A word vectorization unit that converts event information divided for each word into a distributed expression vector for each word by the learned second neural network.
An event vectorization unit that converts the input event information into a distributed expression vector in units of event information based on the distributed expression vector for each word included in the event information.
A vector expression storage unit that stores the event in the storage unit as a distributed expression vector is further provided.
The similarity calculation unit determines the cosine similarity between the event information vectorized by the event vectorization unit and the distributed expression vectorized event stored in the vector expression storage unit. The troubleshooting system according to claim 6, wherein the similarity is calculated based on the above. The troubleshooting according to claim 7, wherein the distributed representation vector corresponding to the meaning of the input word is based on the distributed representation vector of the subword extracted by dividing the input word into N-grams. system.

Images