diff --git a/torch_npu/profiler/analysis/_npu_profiler.py b/torch_npu/profiler/analysis/_npu_profiler.py
index cc23309bb0f90330bb1747df5159824e5ab55b91..2cee702bc512bcd41cc50845c0f77c5576fb0d9e 100644
--- a/torch_npu/profiler/analysis/_npu_profiler.py
+++ b/torch_npu/profiler/analysis/_npu_profiler.py
@@ -1,9 +1,8 @@
 import multiprocessing
 
-
 from .prof_common_func._constant import Constant, print_error_msg
-from .prof_common_func._multi_process_pool import MultiProcessPool
 from .prof_common_func._path_manager import ProfilerPathManager
+from .prof_common_func._multi_process_pool import MultiProcessPool
 from ._profiling_parser import ProfilingParser
 from ...utils._path_manager import PathManager
 
@@ -34,11 +33,21 @@ class NpuProfiler:
                              len(profiler_path_list))
         MultiProcessPool().init_pool(process_number)
 
+        futures = []
         for profiler_path in profiler_path_list:
             PathManager.check_directory_path_writeable(profiler_path)
             profiling_parser = ProfilingParser(profiler_path, analysis_type, output_path, kwargs)
-            MultiProcessPool().submit_task(profiling_parser.analyse_profiling_data)
+            future = MultiProcessPool().submit_task(profiling_parser.analyse_profiling_data)
+            futures.append(future)
         if not kwargs.get("async_mode", False):
+            timeout = 30
+            if timeout:
+                import concurrent.futures
+                done, not_done = concurrent.futures.wait(futures, timeout=timeout)
+                if not_done:
+                    print(f"Profiling analysis timed out after {timeout} seconds")
+                    MultiProcessPool().close_pool(wait=False)
+                    return
             MultiProcessPool().close_pool()
 
     @classmethod
diff --git a/torch_npu/profiler/analysis/prof_common_func/_multi_process_pool.py b/torch_npu/profiler/analysis/prof_common_func/_multi_process_pool.py
index 3f1031a9b3cf0626b31ace2ef41e0f06f221a849..5a3d58e91d821f88d5de8f2c604c9385185f8c41 100644
--- a/torch_npu/profiler/analysis/prof_common_func/_multi_process_pool.py
+++ b/torch_npu/profiler/analysis/prof_common_func/_multi_process_pool.py
@@ -1,4 +1,6 @@
 import atexit
+import os
+import signal
 from concurrent import futures
 
 from ._singleton import Singleton
@@ -8,17 +10,38 @@ from ._singleton import Singleton
 class MultiProcessPool:
     def __init__(self):
         self._pool = None
+        self._processes = set()
         atexit.register(self.close_pool)
 
     def init_pool(self, max_workers=4):
         if not self._pool:
             self._pool = futures.ProcessPoolExecutor(max_workers=max_workers)
+            # 捕获进程创建事件以跟踪PID
+            self._pool._mp_context.Process = self._track_process
         return self._pool
 
+    def _track_process(self, *args, **kwargs):
+        process = futures.ProcessPoolExecutor._mp_context.Process(*args, **kwargs)
+        self._processes.add(process)
+        return process
+
     def close_pool(self, wait: bool = True):
         if self._pool:
             self._pool.shutdown(wait=wait)
             self._pool = None
+        # 强制终止所有未退出的子进程
+        for process in self._processes:
+            if process.is_alive():
+                try:
+                    os.kill(process.pid, signal.SIGTERM)
+                    # 等待进程终止
+                    process.join(timeout=5)
+                    # 如果仍未终止则强制杀死
+                    if process.is_alive():
+                        os.kill(process.pid, signal.SIGKILL)
+                except (OSError):
+                    pass
+        self._processes.clear()
 
     def submit_task(self, func, *args, **kwargs):
         if not self._pool: