A Self-Learning Sensor Fault Detection Framework for Industry Monitoring IoT

Algorithm 1

A group-based fault detection Algorithm.

(1) let 𝑊^𝑚 be the statistics sliding windows size;

(2) let 𝜀_𝑠 be the outlier detection threshold for sensor S and let P be the

  global list to keep all of the 𝜀_𝑠;

(3) let U and V be the global arrays to keep the last 10 gaussian

  distribution characteristics for each sensor;

(4) let 𝑊^𝑛 be the status transform windows size and 𝜃_𝑖/4 be the trend

  vector merging threshold for group 𝐺_𝑖, all the rational trend vectors

  of 𝐺_𝑖 are stored in 𝑅_𝑖;

(5) procedure GFD  ()

(6)  init P, loading 𝜀 from Appilcation DB;

(7)  init U and V, loading the last 10 distribution characteristics from

  Application DB;

(8)  create C DetectionThread threads, 𝐶=𝑁𝑢𝑚_𝑐𝑝𝑢∗2;

(9)  start SelfLearningThread();

(10)  do

(11)   get a value set 𝑣_𝑖 for sensors in a group 𝐺_𝑖;

(12)   find a idle DetectionThread;

(13)    DetectionThread(𝑣_𝑖);

(14)  while not end;

(15) return;

(16) procedure DetectionThread(𝑣_𝑖)

(17)   if (IsStuck(𝑣_𝑖) or IsSpikes(𝑣_𝑖))

(18)     return;

(19)   if (IsOutlier(𝑣_𝑖) and not IsRatStatChange(𝑣_𝑖))

(20)     return;

(21)   IsRatStatChange(𝑣_𝑖);

(22)  return;

(23) procedure IsStuck(𝑣_𝑖)

(24)  get 𝜎²_𝑖𝑗 for sensor j by 𝑇_𝑐;

(25)  if (𝜎²_𝑖𝑗=0)

(26)    mark 𝑠_𝑖𝑗 as a Stuck;

(27) return;

(28) procedure IsSpikes(𝑣_𝑖)

(29)  get 𝜇_𝑖𝑗 and 𝜎²_𝑖𝑗 for sensor j by 𝑇_𝑐;

(30)  if (𝜇_𝑖𝑗/mean(𝑈(𝜇_𝑖))<𝜉 and 𝜎²_𝑖𝑗/mean(𝑉(𝜎_𝑖²))>𝜏)

(31)    mark 𝑠_𝑖𝑗 as a Spikes;

(32) return;

(33) procedure IsOutlier(𝑣_𝑖)

(34)  use U and V to caculate 𝜃_𝑖𝑗;

(35)  if (𝜃_𝑖𝑗>𝜀_𝑖𝑗) mark 𝑠_𝑖𝑗 as an Outlier;

(36) return;

(37) procedure IsRatStatChange(𝑣_𝑖)

(38)  use the last 𝑛 values of 𝑠_𝑖 to fit the trend vector 𝑆_𝑖;

(39)  for each rational trend 𝑟_𝑖 in 𝑅_𝑖

(40)     if (cosine(𝑟_𝑖,𝑆_𝑖)<𝜃_𝑖/4)

(41)       mark 𝑆_𝑖 as a rational status change; 𝑆_𝑖𝑗=mean(𝑆_𝑖𝑗,𝑟_𝑖𝑗);

(42)  mark the max unstable sensor 𝑠_𝑖𝑗 as a sensor falut;

(43) return;

(44) procudure SelfLearningThread()

(45)  while (true) do:

(46)  load the user feedback;

(47)  for each miss alart

(48)     if missed Outlier then 𝜀_𝑖𝑗=𝜀_𝑖𝑗÷𝜆;

(49)     if missed irrational status change then 𝜃_𝑖=𝜃_𝑖×𝜆;

(50)      IncClustering(𝑆_𝑖, null);

(51)  for each false alart

(52)     if false Outlier then 𝜀_𝑖𝑗=𝜀_𝑖𝑗×𝜆;

(53)     if false irrational status change then 𝜃_𝑖=𝜃_𝑖÷𝜆;

(54)     IncClustering(𝑆_𝑖,𝑟_𝑖𝑗);

(55)  sleep(𝑇); continues;

(56) return;

(57) procedure IncClustering(𝑆_𝑖,𝑟_𝑖𝑗)

(58)   if 𝑟_𝑖𝑗 is not null then for each 𝑆_𝑖𝑗 in 𝑆_𝑖;

(59)    if there is cosine(𝑟_𝑖,𝑆_𝑖)<𝜃_𝑖 then 𝑆_𝑖𝑗=mean(𝑆_𝑖𝑗,𝑟_𝑖𝑗);

(60)    else add 𝑟_𝑖𝑗  into 𝑆_𝑖;

(61)  for each two 𝑆_𝑖𝑗, 𝑆_𝑖𝑘 in 𝑆_𝑖;

(62)     if (cosine(𝑆_𝑖𝑗,𝑆_𝑖𝑘)<𝜃_𝑖)

(63)     𝑆_𝑖𝑗=mean(𝑆_𝑖𝑗, 𝑆_𝑖𝑘); remove 𝑆_𝑖𝑘;

(64) return;