CN115144474A - Ultrasonic signal data quality detection method - Google Patents

Abstract

Historical data of ultrasonic detection data of each detection point in a detection interval are collected, a prediction ultrasonic numerical value of the detection point is calculated, real ultrasonic data in the detection interval are obtained, and when the number of data abnormal windows in the detection interval is larger than a set first judgment threshold value, suspected abnormal behaviors of the data are judged to occur. And when the suspected abnormal behavior of the data occurs in the detection interval, carrying out damage inspection on the position of each abnormal window, and when the damage is detected, modifying the attribute of the detection position to be normal. Counting the number of abnormal windows in the processed detection interval, and judging that the data of the detection interval is abnormal when the number is greater than a set first judgment threshold; otherwise, judging that the detection interval data is normal.

Description

Ultrasonic signal data quality detection method

Technical Field

The invention relates to the field of rail transit and steel rail flaw detection, in particular to a method for detecting the quality of steel rail flaw detection ultrasonic signal data.

Background

In recent years, with the increase of railway logistics, the density of trains is improved, the railway load is gradually increased, the abrasion of steel rails is accelerated, and the health state of the steel rails seriously influences the safety of railway traffic, so that the detection work of the steel rails is particularly important. The prior flaw detection work mainly adopts an ultrasonic steel rail flaw detection vehicle to detect the steel rail, collects ultrasonic data and brings the data back to a workshop to complete data playback and analysis work. After the analyst carries out basic judgment on the damage, the damage data are uploaded to the corresponding station segment, and the station segment analyst carries out one-time rechecking; after the analysis of the station analyst is finished, the station analyst submits the data to a group company business department for secondary damage rechecking, and whether the acquired data is accurate or not effectively influences the subsequent working links. In the process of ultrasonic flaw detection, the conditions of inaccurate detection data and the like caused by aging failure of a probe, uneven surface of a rail surface, irregular operation of a flaw detector and the like easily occur, the efficiency of data playback is reduced, health judgment on a steel rail is influenced, conditions of missed judgment, erroneous judgment and the like easily occur, and traffic safety is threatened. Therefore, how to ensure the accuracy of the ultrasonic signal data is particularly critical in the whole rail flaw detection work. The method can more accurately analyze the wave-emitting characteristics of the ultrasonic signals by utilizing the texture analysis method, thereby automatically screening the ultrasonic signal intervals with abnormal wave-emitting, making up the blank of ultrasonic signal quality detection in the current flaw detection field and improving the flaw judgment accuracy.

Disclosure of Invention

In view of the above, the present invention provides a method for detecting quality of ultrasonic signal data for rail flaw detection, which is directed to solve the problems existing in the conventional ultrasonic rail flaw data acquisition process.

The purpose of the invention is realized by the following technical scheme.

A method for detecting the data quality of an ultrasonic signal comprises the following steps:

(1) Collecting historical data of ultrasonic detection data of each detection point in the detection interval, and collecting a set { s } of historical data at any point i _i，j J represents the serial number of the historical data, and the element number of the set is n _i ，m _i ＝int(n _i /3), where int () is a rounding function;

(2) Calculating predicted ultrasonic values of detection points

w _i ＝c ₁ g ₁ +(1-c ₁ )g ₂

Wherein c is ₁ First correction constant for historical data training

g ₁ ＝k ₁ (n _i +1)+b ₁

g ₂ ＝k ₂ (m _i +1)+b ₂

Obtaining a prediction data set { w) in a detection interval _i }；

(3) Acquiring real ultrasonic data s in detection interval _i Get a length N ₁ Time window, traversing the detection point data according to fixed length s, and calculating the difference d between the real data and the predicted data of each detection point in each step of the time window _i ＝|s _i -w _i I, adding to obtain a window difference value in the time window

Where k is the starting sequence number of the window, when D _k Is greater than a set first judgment threshold value ts ₁ And judging that the window k is a data abnormal window. When the number of the data abnormal windows in the detection interval is larger than a set first judgment threshold ts ₂ Judging that the data is suspected to have abnormal behavior;

(4) And when the suspected abnormal behavior of the data occurs in the detection interval, carrying out damage inspection on the position of each abnormal window, and when the damage is detected, modifying the attribute of the detection position to be normal. Counting the number of abnormal windows in the processed detection interval, and when the number is larger than a set first judgment threshold ts ₂ Judging that the detection interval data is abnormal; otherwise, judging that the detection interval data is normal.

Detailed Description

The invention relates to an ultrasonic signal data quality detection method, which comprises the steps of (1) collecting historical data of ultrasonic detection data of each detection point in a detection interval, and aiming at a historical data set { s) at any point i _i，j J represents the serial number of the historical data, and the element number of the set is n _i ，m _i ＝int(n _i /3), where int () is a rounding function;

(2) Calculating predicted ultrasonic values of detection points

w _i ＝c ₁ g ₁ +(1-c ₁ )g ₂

Wherein c is ₁ First correction constant for historical data training

g ₁ ＝k ₁ (n _i +1)+b ₁

g ₂ ＝k ₂ (m _i +1)+b ₂

Obtaining a prediction data set { w) in a detection interval _i }；

(3) Acquiring real ultrasonic data s in detection interval _i Get a length N ₁ Time window, traversing the detection point data according to fixed length s, and calculating the difference d between the real data and the predicted data of each detection point in each step of the time window _i ＝|s _i -w _i I, adding to obtain a window difference value in the time window

Where k is the starting sequence number of the window, when D _k Is greater than a set first judgment threshold value ts ₁ And judging that the window k is a data abnormal window. When the number of the data abnormal windows in the detection interval is larger than a set first judgment threshold ts ₂ Judging that the data is suspected to have abnormal behavior;

(4) And when the suspected abnormal behavior of the data occurs in the detection interval, carrying out damage inspection on the position of each abnormal window, and when the damage is detected, modifying the attribute of the detection position to be normal. Counting the number of abnormal windows in the processed detection interval, and when the number is larger than a set first judgment threshold ts ₂ Judging that the detection interval data is abnormal; otherwise, judging that the detection interval data is normal.

Claims (1)

1. A method for detecting the data quality of an ultrasonic signal is characterized by comprising the following steps:

(1) Collecting historical data of ultrasonic detection data of each detection point in the detection interval, and collecting a set { s } of historical data at any point i _i，j J represents the serial number of the historical data, and the element number of the set is n _i ，m _i ＝int(n _i /3), where int () is a rounding function;

(2) Calculating predicted ultrasound values for detection points

w _i ＝c ₁ g ₁ +(1-c ₁ )g ₂

Wherein c is ₁ First correction constant for historical data training

g ₁ ＝k ₁ (n _i +1)+b ₁

g ₂ ＝k ₂ (m _i +1)+b ₂

Obtaining a prediction data set { w) in a detection interval _i }；

(3) Acquiring real ultrasonic data s in detection interval _i Get a length N ₁ Time window, traversing the detection point data according to fixed length s, and calculating the difference d between the real data and the predicted data of each detection point in each step of the time window _i ＝|s _i -w _i I, adding to obtain a window difference value in the time window

Where k is the starting sequence number of the window, when D _k Is greater than a set first judgment threshold value ts ₁ And judging that the window k is a data abnormal window. When the number of the data abnormal windows in the detection interval is larger than a set first judgment threshold ts ₂ Judging that the data is suspected to have abnormal behavior;

(4) And when the suspected abnormal behavior of the data occurs in the detection interval, carrying out damage inspection on the position of each abnormal window, and when the damage is detected, modifying the attribute of the detection position to be normal. Counting the number of abnormal windows in the processed detection interval, and when the number is larger than a set first judgment threshold ts ₂ Judging that the detection interval data is abnormal; otherwise, judging that the detection interval data is normal.